WO2000070528A2 - Method and apparatus for predictive cellular bioinformatics - Google Patents

Abstract

Techniques for using information technology in therapeutics or drug discovery. In an exemplary embodiment, techniques for determining information about the properties of substances based upon information about structure of living or non-living cells exposed to substances are provided. A method according to the present invention enables researchers and/or scientists to identify promising candidates in the search for new and better medicines or treatments using, for example, a cellular informatics database. The present invention further teaches a system for acquiring knowledge from cellular information. The system has a database 1012 comprising a database management module ('DBMS'). The system also has a variety of modules, including a population module coupled to the DBMS for categorizing and storing a plurality of features (e.g., cell size, distance between cells, cell population, cell type) from an image acquisition device into the database. The system has a translation module coupled to the DBMS for defining a descriptor from a set of selected features from the plurality of features. In a specific embodiment, the descriptor is for a known or unknown compound, e.g., drug. A prediction module is coupled to the DBMS for selecting one of a plurality of descriptors from known and unknown compounds from the database based upon a selected descriptor from a selected compound. The selected compound may be one that is useful for treatment of human beings or the like.

Description

PATENT APPLICATION

METHOD AND APPARATUS FOR

PREDICTIVE CELLULAR BIOINFORMATICS

COPYRIGHT NOTICE A portion ofthe disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone ofthe patent document or the patent disclosure as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever. The present description includes examples of computer codes, which may be used to implement aspects ofthe present invention. Assignee ofthe present invention reserves all rights with respect to these codes and provides notice herein. Notice is hereby given © Cytokinetics, Inc. 1999.

BACKGROUND OF THE INVENTION The present invention provides techniques for information management using a database platform. More particularly, the present invention provides a system including computer code that couples to a database device. The system provides for image capturing of living, dead, or fixed cells or cell fractions used to identify information about substances used on the cells or information about the cells themselves. Accordingly, the present invention can enable researchers and scientists to identify promising candidates in the search for new and better medicines, for example, in drug discovery and development. The principles enumerated herein may, with equal facility, be applied to other applications, including but not limited to use in environmental applications such as determining chemical toxicities and other non-pharmaceutical toxicology uses. For a long time, researchers in the pharmaceutical field have sought for better ways of searching for substances possessing properties that make them suitable as medicines. In the early days, researchers generally relied upon extracts from plants, dyes, and microbiological extracts for such substances. Examples of such substances include the pain reliever aspirin, the anti-cancer drug paclitaxel (brand name Taxol™), and the heart medication called digoxin. The number of useful medicines has generally been limited. Purified substances having desirable bio-active properties are also often difficult to discover. Advances in traditional organic chemistry and more recently the rapid chemical synthesis methods often referred to as combinatorial chemistry have increased the number of compounds that researchers test for biological activity. Originally, substances were often initially tested on animals or humans to determine their biological activity. While results from such tests may identify a good drug candidate, they are often time consuming and costly, thus a limited number of substances can be tested. Therefore, pharmaceutical companies have turned to testing their ever-increasing libraries of substances against isolated proteins (drug targets) in biochemical assays that can be carried out at high throughput and low cost. It should be noted that the substances need to be tested in numerous protein tests, each customized for a particular drug target. Therefore, although each protein test may be run at a high-throughput, the design of multiple protein tests can be time-consuming. Substances deemed promising based on results from the protein tests are then tested in lower throughput cellular and animal tests.

There have been some attempts to use image acquisition techniques to screen a large number of substances based upon biological cell information. One such attempt is described in International Application No. WO 98/38490 in the names of Dunlay, et al. Dunlay et al. generally describes a conventional image acquisition system. This conventional system collects and saves images based on certain criteria that are predefined, not on a fixed area of an imaging surface. Additionally, the conventional system has poor lighting design, which makes image processing for multiple cells difficult. Furthermore, the conventional system is not designed for capturing, populating and utilizing a large database design. The conventional system is designed for customized cellular assays, not as a tool for generation of a cellular informatics database. Without such database capabilities the conventional system cannot be used for screening, analyzing, and comparing large quantities of cells from multiple experiments on multiple days in a predictive, efficient and cost effective manner. What is needed is a rapid assay to assess the activity of compounds against multiple drug targets simultaneously in a cellular context. What is also needed are techniques for finding the effects of substances on cell function based upon searching and analyzing cellular information. SUMMARY OF THE INVENTION According to at least one embodiment of the present invention, techniques for determining information about effects of potential substances on cells are provided. In another exemplary embodiment, the present invention provides a novel system including hardware, computer codes, user interfaces, and a database for acquiring, storing and retrieving cellular and substance information. The cells can include living, dead, or fixed cells or fractions of cells. The present invention enables, inter alia, researchers and/or scientists to identify promising candidates in the search for new and better medicines or treatments using, for example, a cellular informatics database.

According to the present invention, a computer program for identification and verification of biological properties of substances can include code that causes a sample of a substance to be administered to a cell. The code determines one or more features for two or more cell components, or markers, in the presence of the substance. The code can form one or more descriptors from the features. Descriptors can be formed by combining features of two or more cell components as identified using the markers. The code can then search one or more descriptors obtained from prior administered substances upon cells in order to locate descriptors having a relationship to the descriptors noted for the substance under study. The code predicts properties ofthe administered substance based upon the properties ofthe prior administered substances using the relationship between the descriptors. The code can provide for identifying properties of substances based upon effects on cell characteristics. Candidate drug mechanisms of action, potency, specificity, pharmacodynamic, and pharmacokinetic parameters, toxicity, and the like can be used as substance properties.

In a specific embodiment, the present invention provides a system for acquiring knowledge from cellular information. The system has a database comprising a database management module ("DBMS")- The system also has a variety of other modules, including a population module that is coupled to the DBMS and serves to categorize and store a plurality of features (including but not limited to cell size, distance between cells, cell population, as well as sub-cellular features such as organelle location, protein location and sub-cellular constituent location and movement) from an image acquisition device into the database. The system has a translation module coupled to the DBMS for defining a descriptor from a set of selected features from the plurality of features. In a specific embodiment, the descriptor is for a known or unknown compound, e.g., drug. A prediction module is coupled to the DBMS for selecting one of a plurality of a descriptors from known and unknown compounds from the database based upon a selected descriptor from a selected compound. The selected compound may be one that is useful for treatment of human beings or the like.

In a specific embodiment, the present invention provides a system for populating a database with cellular information. The system includes a cell holder (e.g., multi-well plate, chip, micro fluidic assembly, or other cell chamber) comprising a plurality of sites in a spatial orientation. Each of the sites is capable of holding a plurality of cells to be imaged. Note - the light guide is one embodiment, but we don't want to be limited to it. According to one embodiment, the present system also has an illumination apparatus including a liquid light guide operably coupled to the imaging device for highlighting the plurality of cells in a relatively even spatial manner for image capturing and measurement purposes. Still further, the liquid light guide allows sub-elements (e.g., filter, lamp) ofthe illumination apparatus to be placed at a remote location to prevent mechanical interference of the cell holder during image capturing. Alternative lighting methodologies may, with equal facility, be implemented.

The system also has an image-capturing device (e.g., charge coupled device camera, translation stage, shutter, microscope, software, shutter control) coupled to a computing device (e.g., computer, network computer, work station, analog computing device, on-board image-processor, and laptop). The image-capturing device is adapted to capture at least one image in at least one ofthe plurality of sites. One some embodiments, multiple images can be captured, where each image represents a different cell component (or portion). The image-capturing device can be adapted to convert the image into a digital representation, which highlights the feature or features of the one site.

A database storage device (e.g., relational database, object oriented database, mixed object oriented database) includes a database management element. The database is coupled to the image capturing device. In a specific embodiment, the present system includes modules for feature extraction, generation of descriptions, and data preparation and analysis.

In a specific embodiment, the present invention provides a novel system for determining an effect of a manipulation of a cell using one or more image frames. The system has a plate comprising a plurality of sites in a spatial orientation. Each ofthe sites is capable of holding a plurality of cells to be imaged. The system also has an image capturing device to capture a plurality of images of at least one site from the plurality of sites. The image capturing device is coupled to the computing device. The system also has an image processing device to combine the plurality of images of at least one site or plurality of sites. The image processing device is operably coupled to the plate. An image processing device is also included. The image processing device can be adapted to form a digitized representation ofthe plurality of images from the site or plurality of sites. Furthermore, the system has a database storage device comprising a database management element. The database can be adapted to retrieve the descriptor or descriptors ofthe plurality of features from the computing processing device and storing them in a selected manner.

In a specific embodiment, the present invention provides a system for capturing cellular information. The system also has an image acquisition system comprising a charged coupled device camera adapted to capture an image of a plurality of manipulated cells in various stages of the cell cycle. The stages ofthe cell cycle are currently understood to include interphase, GO phase, Gl phase, S phase, G2 phase, M phase, prophase, prometaphase, metaphase, anaphase, and telophase. The principles ofthe present invention specifically contemplate the application thereof on additional cell cycle stages when and if they are identified.

An optical source is coupled to the image acquisition system for highlighting the plurality of manipulated cells in the various stages ofthe cell cycle. The illumination apparatus provides for an acquisition ofthe image ofthe plurality of manipulated cells. In a specific embodiment, the illumination apparatus has a liquid light guide coupled to a light source at a remote location.

A variety of user interfaces are utile for accessing the several features ofthe present invention. Those having ordinary skill in the art will appreciate that different user interfaces may be required to support different research scenarios. The present invention specifically contemplates the utilization of a wide variety of user interfaces.

Numerous benefits are achieved by way ofthe present invention over conventional techniques. The present invention can provide techniques for predictive cellular bioinformatics that can streamline a number of important decisions made in the drug discovery industry. The present invention can be implemented using off the shelf hardware including databases. In other aspects, the present invention can find useful information about substances as well as cells or portions of cells. Furthermore, the present invention can acquire more then one feature using more than one manipulation. Moreover, the present invention can provide information about a wide variety of cellular information that is not conventionally available. This information includes information about different cell components, e.g., nuclei and Golgi apparatus. Still further, the present invention provides an automated or semi- automated technique for acquiring images and populating a database. The present database can be combined with others such as genomics, and the like. Moreover, the present invention can be implemented to predict, ter alia, a mechanism of action, toxicity, target validation, and pre-clinical disease model.

A further understanding ofthe nature and advantages ofthe invention herein may be realized by reference to the remaining sections ofthe specification and the attached drawings.

BRIEF DESCRIPTION OF THE DRAWING

For more complete understanding of the present invention, reference is made to the accompanying Drawing in the following Detailed Description of the Invention. In the drawing:

Fig. 1 is a simplified system diagram according to an embodiment according to the present invention;

Figs. 1A-1B are more detailed diagrams of database systems according to embodiments ofthe present invention;

Fig. 2 is a simplified block diagram according to an alternative embodiment according to the present invention;

Figs. 3-6 are simplified diagrams of system elements according to embodiments of the present invention; Figs. 7A-7K illustrate representative block diagrams of simplified process steps in a particular embodiment according to the present invention;

Fig. 8A-8F illustrate representative quantified descriptors of effects of manipulations on images of cells in a particular experiment;

Fig. 9 illustrates example images for different types of morphologies in a particular experiment;

Fig. 10 illustrates a distribution of various morphologies in a cell population responsive to drug concentration in a particular experiment;

Fig. 11 illustrates a graph of quantified features of effects of manipulations on cells in a particular experiment; Fig. 12 illustrates effects of external agents on cells in a particular experiment;

Fig. 13 illustrates 4 panels for each marker for a plurality of A549 cells in a particular experiment;

Fig. 14 illustrates 4 panels for each marker for a plurality of OVCAR-3 cells in a particular experiment;

Fig. 15 illustrates 4 panels for each marker for a plurality of OVCAR-3 cells at 20x in a particular experiment; Fig. 16 illustrates 4 panels for each marker for a plurality of OVCAR-3 cells at 40x in a particular experiment;

Fig. 17 illustrates a representative input for a morphometric analysis program in a particular embodiment according to the present invention; and Figs. 18-19 illustrate examples of the generation of pseudo-sequences and clustering in a particular embodiment according to the present invention.

Fig. 20 is a block diagram for a first research scenario;

Fig. 21 is a block diagram for a second research scenario; and

Fig. 22 is a block diagram for a third research scenario. Reference numbers refer to the same or equivalent parts of the invention throughout the several figures ofthe Drawing.

DETAILED DESCRIPTION OF THE INVENTION According to the present invention, techniques for determining information about manipulated cells or substances based upon living, fixed, or dead cell structures or portions of cells are provided. In an exemplary embodiment, the present invention provides a novel system including computer codes coupled to a database and user interfaces for acquiring, storing and retrieving such information. Other embodiments provide a novel image capturing system for providing digitized representations of live and dead cell structures or the like. Fig. 1 is a simplified system diagram 10 of a cellular knowledge-based system according to an embodiment ofthe present invention. This diagram is merely an example and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The present system 10 includes a variety of elements such as a computing device 13, which is coupled to an image processor 15 and is coupled to a database 21. The image processor receives information from an image capturing device 17, which image processor and image capturing device are collectively referred to as the imaging system herein. The image capturing device obtains information from a plate 19, which includes a plurality of sites for cells. These cells can be biological cells that are living, fixed, dead, cell fractions, cells in a tissue, and the like. The computing device retrieves the information, which has been digitized, from the image processing device and stores such information into the database. A user interface device 11 , which can be a personal computer, a work station, a network computer, a personal digital assistant, or the like, is coupled to the computing device. Fig. IA is a simplified diagram of a database system 1000 according to an embodiment ofthe present invention. This diagram is merely an example and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize many other variations, modifications, and alternatives. Database system 1000 includes a variety of techniques for processing images from biological cells, e.g., fixed, living, and dead cells, and cell portions. As shown, images are acquired 1001. These images can be from a single frame or multiple frames. As merely an example, an image processing system may analyze such images. One example of such an image processing system is described below, but should not be construed as limiting ceπain claims.

In a specific embodiment, cell samples are manipulated using a compound (e.g., substance, drug). The cell samples are imaged for a simple portion or portions, e.g., manipulated cell substructure, manipulated spatial feature of cell, cell density. Image processing techniques are used to extract 1003 the feature or features from the image or images. The features can be an independent or a dependent set of cell characteristics (which may be predominately visual) including, for example, count, area, perimeter, length, breadth, fiber length, fiber breadth, shape factor, elliptical form factor, inner radius, outer radius, mean radius, equivalent radius, equivalent sphere volume, equivalent prolate volume, equivalent oblate volume, equivalent sphere surface, average intensity, total intensity, optical density, radial dispersion, texture difference, and others. Each of these features corresponds to a similar manipulation by a compound. Each manipulation forms a new set of features, which are identifiable to the compound. Once each set of features has been extracted, the feature set is populated 1004 into a database 1012. Accordingly, the database includes many sets of features, where each set corresponds to a different manipulation for a selected cell. Each set of features corresponding to a manipulation provides a descriptor 1009, which is also stored 1019 in the database. The descriptor is a "finger print" including each feature for the manipulation. Each descriptor may be unique, or may have similarities to other descriptors or may even be the same as other descriptors for known and unknown manipulations.

The present system retrieves features, which we define as simple features herein, and forms composite features 1007 from them. More than one feature can be combined in a variety of different ways to form these composite features. In particular, the composite feature can be any function or combination of a simple feature and other composite features. The function can be algebraic, logical, sinusoidal, logarithmic, linear, hyperbolic, statistical, and the like. Alternatively, more than one simple feature can be combined in a functional manner (e.g., arithmetic, algebraic). As merely an example, the composite feature equals a sum of feature 1 and feature 2, where these features correspond to the same manipulation. Alternatively, the composite feature equals feature 1 divided by feature 2. Alternatively, the composite feature equals feature 1 minus feature 2. Alternatively, the composite feature equals a constant times feature 1 plus feature 2. Of course, there are many ways that the composite feature can be defined. The present system also stores 1017 these features in the database. The composite features can also be further combined with simple features. Once these features are defined as descriptors, they are stored 1019 in the database.

Fig. IB is a simplified diagram of a database system engine 2000 according to an embodiment of the present invention. This diagram is merely an example and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize many other variations, modifications, and alternatives. The engine can be implemented into the present database for populating, searching, and predicting compound or cell characteristics. As merely an example, engine 2001 includes an input/output module 2008. The input/output module is used to input and output information from the database. The information includes, among others, a plurality of feature sets, which correspond to many manipulations. Additionally, the information includes descriptors, which each corresponds to a set of features from the manipulation. The database also has a population module, which is used to configure the features based upon an entity relationship, which has been predetermined. The database engine also has other modules. In particular, the database has a transcription module, which transfers a preselected set of features and creates a descriptor from them. The transcription module can be used to take a known compound, which has features, to transcribe them into a descriptor. Altematively, the transcription module can be used to take an unknown compound, which has features, to transcribe them into a descriptor. These descriptors are provided into the database for subsequent use. Finally, the database engine has a prediction module, which can be used to potentially predict a property (e.g., mechanism of action) of an unknown compound. Here, the unknown compound is provided with a descriptor, but the property ofthe compound is unknown. In one embodiment, the prediction module compares a descriptor of an unknown compound with the many descriptors of known compounds, which were in the populated database. Depending upon the matching criteria, the prediction module will attempt to uncover one or more descriptors of known compounds. Once the prediction module finds the descriptors ofthe known compounds based upon the descriptor for the unknown compound, it identifies a potential property of such unknown compound for analysis and review. Here, it is believed that certain features ofthe known compound, which are similar to those features ofthe unknown compound may uncover a property to the unknown compound. Details ofthe present software engine are described more fully below. Fig. 2 is a simplified block diagram 20 of a cellular knowledge-based system according to an alternative embodiment of the present invention. This diagram is merely an example and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Like reference numerals are used in the present diagram as the previous diagram for easy cross-referencing, but are not intended to be limiting in any manner. The present diagram 20 includes a variety of elements such as a processor 13 or computing device coupled to a database 11. The processor can be used for retrieving and storing information from the database. The system also includes a plurality of system elements, such as a cleaner 23, a dispenser 25, and an image capturing system 27, which are also coupled to the database in some embodiments. These elements can be coupled to each other through a network or the like. As merely an example, the network can be a NetWare™ network from Novell Corporation or an internet network or the Internet but can also be others and any combination thereof. The system also has an output device 31, which can be used to output information from the database, processor, or other system elements. Details of these elements are described more fully below in reference to the Figs.

Figs. 3-5 are simplified drawings of system elements according to embodiments ofthe present invention. These diagrams are merely examples and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. As merely an example, Fig. 3 is a simplified diagram of a processor or computing device 13. The computing device 13 includes a bus 1 12 which interconnects major subsystems such as a central processor 114, a system memory 1 16 (e.g., random access memory), an input/output ("I/O") controller 118, an external device such as a display screen 124 via a display adapter 126, a keyboard 132 and a mouse 146 via an I O controller 118, a SCSI host adapter (not shown), and a floppy disk drive 136 operative to receive a floppy disk 138.

The computing device has other features. Storage Interface 134 may act as a storage interface to a fixed disk drive 144 or a CD-ROM player 140 operative to receive a CD-ROM 142. Fixed disk 144 may be a part of computing device or may be separate and accessed through other interface systems. A network interface 148 may provide a direct connection to a remote server via a telephone link or to the Internet. Network interface 148 may also connect to a local area network ("LAN") or other network interconnecting many computer systems. Many other devices or subsystems (not shown) may be connected in a similar manner. Also, it is not necessary for all of the devices shown in Fig. 3 to be present to practice the present invention, as discussed below. The devices and subsystems may be interconnected in different ways from that shown in Fig. 3. The operation of a computer system such as that shown in Fig. 3 is readily known in the art and is not discussed in detail in this application. Computer code to implement the present invention, may be operably disposed or stored in computer-readable storage media such as system memory 116, fixed disk 144, CD-ROM 140, or floppy disk 138. The computer code can be organized in terms of processes or modules, depending upon the application. That is, the computer code can include a prediction module, a translation, module, or other modules to carryout the functionality described herein, as well as others.

Figs. 4 and 5 are simplified diagrams of an imaging system 200 according to an embodiment ofthe present invention. As shown, the imaging system 200 includes a variety of features such as housing 203, which holds a stage assembly 204. The stage assembly includes an x-stage movement element 206, which is along an x-direction, and a y-stage movement element 207, which is along a y-direction. The imaging system also includes a z-direction movement element, which is perpendicular to the x-y plane. The z-direction movement motor can be attached to the stage, or to the objective nosepiece by way ofthe microscope housing, or as an external motor between the objective and the microscope housing. The stage can align in any one ofthe directions to an accuracy of one micron and less, or one-half micron and less, or one-quarter micron and less, depending upon the embodiment.

The stage holds a plate 202 or cell holder, which houses one of a plurality of samples. The plate includes a spatial array 209 of process sites. Each of the process sites can include a plurality of cells and solutions depending upon the embodiment. Each ofthe sites can carry a sufficient amount of solution to prevent substantial evaporation of the sample during processing in some embodiments. In embodiments for large scale analysis, the plate includes at least 96 sites, or more than or equal to 384 sites, or more than or equal to 1,536 sites. The plate bottom is transparent and thin, which allows light to pass through the sample. Additionally, the plate is made of a suitable chemical resistant material. As merely an example, the plate can be either a 96, or 384, or 1536 or other formats from places such as Becton Dickinson of Franklin Lakes, NJ, or Coming Science Products of Coming, NY. In a preferred embodiment, the plate is a Coming Costar black-walled 96 well plate catalog #3904 from Coming Science Products of Co ing, NY, but should not be limited to these in some applications, but can be others.

Also shown is the condenser for the microscope 201 , which can be used to collect phase, DIC, or bright field images of the cells. Images resulting from the illumination ofthe samples to fluorescence, phase, DIC, or bright field techniques are collected using an image capturing device 208, which captures an image or images of cells from the plate. In a specific embodiment, the microscope is an inverted configuration with the objectives on the bottom ofthe plate and the condenser disposed overlying an upper surface ofthe sites, while the image capturing device underlies the sites. Images captured by the imaging device, whether analogue or digital, are viewed by a monitor or other devices. The image capturing device can be any camera assembly such as a charge coupled device camera, which is known as a CCD camera, or other high resolution camera capable of capturing images from the sites. In a specific embodiment, the camera is an interline CCD camera which does not require an external shutter.

In a specific embodiment, the present imaging system can be any suitable unit that is flexible for automated image collection using multi-well plastic plates. The imaging system also should be adapted to collect high-resolution images of cells on plastic or glass plates, cell growth chambers, or coverslips. The system also can be used for imaging multiple cell markers in multiple imaging conditions. To accomplish this, the microscope system has a variety of elements such as a light source, a motorized excitation filter wheel and shutter, x-y-z-motorized stage, excitation and emission filters, Fluor phase and DIC objectives, motorized objective nosepiece, dichroic filters, motorized dichroic filter cubes, phase and DIC rings and prisms, CCD camera, and software control. As merely an example, the present imaging system can have components such as those listed in the Table below.

Table: Image Acquisition System Elements

In a specific embodiment, the present system has the following capabilities, which are not intended to be limiting. Image acquisition

1 ) Ability to automatically acquire multi-wavelength images from multiple sites on one multi-well plate, to sequentially name image files, and to log any imaging parameter information with image files.

2) Ability to link images with a larger database/spreadsheet of information.

3) Ability to automatically collect multiple plates by interfacing the imaging system with a robotic arm.

X-Y control

1) Ability to place 96, 384, or 1536 well plates onto microscope stage and move to each well sequentially. 2) Ability to return to each well and collect another round of images (multi-site time-lapse) or ability to collect rapid time-lapse information at each well (time-lapse of many wells).

3) Ability to collect a low magnification image, automatically determine features which may be of interest, automatically change the objective to a higher magnification, and collect high magnification images of a fixed number of those identified cells in the sample.

4) Ability to collect multiple frames in each site.

Z control

1. Ability to auto-focus with substantially minimal damage to biological specimen or fluorophore.

2. Ability to auto-focus rapidly.

The present embodiment of the imaging system is shown by way of

Figs. 5A and 5B. These diagrams are merely examples and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The present imaging system 40 includes a variety of elements such as a microscope 41, which is preferably an epi-fluorescent microscope, but can be confocal, multiphoton, or hybrid types. The microscope includes elements 41A, the motorized Z-axis; 41B, the motorized dichroic filter cube holder; and 41C, the motorized objective nosepiece. In one embodiment, the microscope is a Model 100M made by Zeiss. The microscope communicates to computer 51 through control lines 73, 75, and 76. The imaging system also has camera 50 coupled to controller 50A and computing device 51 , which oversees and controls operations ofthe elements ofthe imaging system.

The present microscope includes drivers for spatially moving a stage in two dimensions, including an x-direction, a y-direction, and moving the objective nosepiece in a z-direction in a Cartesian coordinate system. The z-direction movement is provided using a fast z-motor, which can make z-direction adjustments within a predetermined time. The z-direction movement generally provides for focussing ofthe sample to the camera. The focussing occurs within the predetermined time of preferably ten seconds and less, or five seconds and less, or one second and less, depending upon the embodiment. As merely an example, the z- motor or positioner can be a model PIFOC objective nanopositioner made by a company called Physik Instrumente of Waldbronn, Germany, but also can be others. The z-motor couples to computer 51 through line 63, which may also include a controller. Depending upon the embodiment, a second z-motor 41 A connected to the computer 51 by line 73 may be used to keep the z-motor 42 in the center of its travel. Alternatively, in other embodiments the stage could be provided with a z-motor allowing for movement of the stage in the z-direction.

The present stage also includes an x-y stage 43. The x-y stage moves plate 59, e.g., 96 site, 384 site, 1536 site. The x-y stage moves plate in an x-y spatial manner. The stage has an accuracy or repeatability of about 1 micron and less, or about 2 microns and less. The stage can move in a continuous manner or a stepped manner. The stage also can move up to 30 mm/sec. or faster. The stage also can move 1 mm/sec. and less, depending upon the embodiment. The stage can also step 0.1 micron and less or 1 micron and less, as well as other spatial dimensions. The stage can be one such as a Proscan Series made by Prior Scientific of Rockland, MA but can also be others. The stage is controlled via control line 61 through controller 43 A, which couples to computer 51 through control line 65.

The stage includes plate holder 44. The plate holder can hold a single plate. In other embodiments, plate holder can also hold multiple plates. The plate holder can use mechanical, electrical, fluid, vacuum and other means for holding the plate or plates. The plate holder also is sufficiently stable for securing the plate. As merely an example, the plate holder is a Model 500-H223R made by Prior Scientific of Rockland, MA . In some embodiments, the plate holder may need adjustment in the z-direction to provide for a desirable focus of a sample on a plate. In these embodiments, the plate holder is supported by spacers 45 or a plurality of stage pins, which mechanically elevate the plate holder in the z-direction. These pins are generally made ofa suitable material for supporting such plate holder and also are sufficiently resistant to chemicals and the like. In some embodiments, the entire imaging system is placed on an isolation table 49. The isolation table is disposed between the microscope and support stmcture. The isolation table is designed to prevent excessive vibration ofthe plate. The isolation table is made of a suitable material such as steel and honeycomb but can be others. The table has a thickness of about 8 inches or preferably less than about 24 inches. In one embodiment, the table is Model 9101-24-85 made by Kinetic Systems of Boston, MA.

The imaging system also has a lamp or illumination assembly 62. The lamp assembly provides for a light source (See reference letter B) to a plurality of elements in the imaging system. For easy reading, the light path is defined by the doted lines, which are not intended to be limiting. The lamp assembly has a variety of elements such as a Xenon lamp 46. The Xenon lamp provides light at about 320 to 700 nanometers (Prefocused). The Xenon lamp is 175 or 300 Watts. As merely an example, the lamp can be a Lambda Model made by Sutter Instmment Company of Novato, CA.

Referring to Fig. 5B, the lamp assembly also has a cold mirror 58, an excitation filter wheel 48, excitation filter(s) 55, and an excitation light shutter 57. As shown, light is derived from the Xenon lamp, reflects off of the cold mirror 58, traverses through the excitation filter or filters 55, and is controlled by the excitation light shutter 57. The lamp assembly has filter wheel 48, which houses one of a plurality of filters, including excitation filters. The shutter and filter wheel are controlled via control lines 67, which are coupled to a computer 51 or other type of computing device. The control lines 67 are coupled through controller 57A (for element 57) and controller 48 A (for element 48) via control line 69 to computer 51. Preferably, light traverses from the lamp assembly through a light guide 47 to illuminate features within the plate. The light guide is suitably selected to have a flexible member, which can be used to place lamp source at a remote location away from the imaging device. The flexible member substantially keeps any vibration from the lamp assembly away from the imaging device. In some embodiments, the member is at least 1 foot away from the imaging device. The light guide is a guide, which is a flexible hose-type sleeve. The sleeve is filled with a liquid such as an aqueous solution containing chloride or phosphate. A thin layer may be formed on the inside ofthe sleeve. The layer can be a containing tetrafluoroethylene and mexafluoropropylene, or containing tetrafluoroethylene and perfluoromethyl vinyl ether, or tetrafluoroethylene and perfluoropropyl vinyl ether. An example of such a light guide is described in International Application No. WO/98/38537 filed February 29, 1997, and assigned to NATH, Gunther. The liquid light guide has less than about 30% transmission loss of the light at a remote location such as the imaging system.

Light is derived from the lamp assembly and directs off of filter 56, which directs the light upward. Filter 56 can be a dichroic and emission filter, as well as others. The light traverses through microscope nosepiece 41 C, and traverses through objective spacers 54. An objective 53 magnifies the light toward a predetermined point on the plate 59. The objective can be, for example, made by Zeiss of Jena, Germany, as well as other companies. The objective can be one of a plurality including IX, 10X, 20X, 40X, and others, depending upon the application. Magnification can be further expanded or contracted by intermediate optics between the objective and the camera. Selection of filter or filters is controlled by computer 51 via control line 75.

The camera 50 captures an image of cells from plate 59. The image is obtained from light scattering off of cells or portions of cells in the plate through objective 53, through objective spacers, through filters 56, which are captured at camera 50. In this preferred embodiment, the camera is a digital camera, but can be an analogue camera. The digital camera is a CCD camera, which has 1280 by 1024 pixels, or more or less. The pixels can be 6.7 microns in dimension or more or less. The camera preferably is substantially free from an external shutter to quickly capture a plurality of images of cells from the plate. The camera is controlled via control line 71 through controller 50A, which connects to computer 51 through control line 70. The present invention can also include other types of image acquisition devices selected from at least an epi fluorescence, a confocal, a total-internal reflection, a phase, a Hoffman, a bright field, a dark field, a differential interference contrast, an interference reflection, or multi-photon illumination device.

The present imaging system stores images on a high density memory device 60. The high density memory device is preferably optical, but can also be magnetic. The high density memory device can be any suitable unit that is capable of storing a plurality of images from a plurality of sites in the plate. The memory device can be a compact disk, which would generally use a compact disk burner or the like. Depending upon the embodiment, the high density memory device is used to archive the images that are captured from the camera in the imaging system. Further details of the imaging system can be found throughout the present specification, and more particularly below.

As merely an example, the present invention can be implemented using the following sequence of steps, which have been described in a journal entry form. Here, images are opened and objects are identified based on a background value that has been edited in starting image acquisition. Information is maintained in a spreadsheet or other database format, which has the following information for each object:

After computations are done, the log file is saved. In particular, the file is saved in an appropriate place with an appropriate name.

In a specific embodiment, the present invention provides the following detailed example of journal entries, which should not limit the scope of the invention.

3X3 IMAGE COLLECTION.ini

Stage (Scan) Takes 9 images of well, -1600 motor steps apart from left to right 3 columns and 3 rows, runs FOCUS, COLLECT IMAGE, SAVE SEQUENTIAL FILE NAME.JNL.

FOCUS. COLLECT IMAGE. SAVE SEOUENTIAL FILE NAME.inl.

START IMAGE ANALYSIS.jnl

IMA OBJECTS.ini

Log obi and sum data.jnl

COLLECT AUTOMATED IMA DATA INONE SPREADSHEET.jnl

OPEN OBJECT LOG DDE FILE.jnl

COLLECT AUTOMATED IMA DATA INONE SPREADSHEET 16 BIT IMAGES. jnl

IMA OBJECTS 16bit.ini

START IMAGE ANALYSIS 16 to 8 BIT.jnl

IMA OBJECTS WITH NEW LOG FILE.jnl

INTERACTIVE IMA OBJECTS.jnl

COLLECT INTERACTIVE IMA DATA.inl

CHANGE FILTER. COLLECT IMAGE. SAVE SEOUENTIAL FILE NAME.inl

COLLECT HOECHST AND FITCinl

3X3 IMAGE COLLECTION HOECHST FITCjnl

AUTOMATED 3X3 IMAGE COLLECTION HOECHST FITCjnl

AUTOMATED IMAGE COLLECTION.ini

STARTUP.ini

Nested Journals

Automated 3x3 Image Collection Loop 3x3 image collection Loop focus, collect image, save sequential file name

Automated 3x3 image collection Hoechst FITC Loop 3x3 image collection Hoechst FITC loop Collect Hoechst and FITC focus, collect image, save sequential file name change filter, collect image, save sequential file name

Automated image collection

Loop focus, collect image, save sequential file name

Collect automated IMA data in one Spreadsheet Open object log DDE file Loop EM A objects Log obj and sum data

Collect automated IMA data in one spreadsheet 16 bit images Open object log DDE file

Loop EMA objects 16 bit Log obj and sum data

Although the above has been generally described in terms of a specific user interface and software code, other user interfaces and code can also be used. One of ordinary skill in the art would recognize many other variations, alternatives, and modifications.

Fig. 6 is a simplified diagram 600 of a cleaning and dispensing system according to an embodiment ofthe present invention. This system 600 includes a variety of elements such as a dispensing head 609, which is coupled to a plurality of pipettes 601. The pipettes input and output fluids or solutions from plate 603. The plate has a plurality of sites, each of which can be used to input cells or a combination of cells and solution. The system also has elements to house solutions 605, which are used to manipulate cell samples in the plate. The dispensing head is supported through a support member 607, which is sufficiently rigid to allow for movement of the head. The dispenser is coupled to the present system in a mechanical and electrical manner, which provides for a fully integrated system for providing cell samples to the imaging system according to the present invention.

Fig. 7 A illustrates a representative block flow diagram of simplified process steps of a method for determining properties of a manipulation based upon effects ofthe manipulation on one or more portions of one or more cells in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. In step 700, one or more samples of cells can be provided. These cells can be live, dead, or fixed cells, or cell fractions. The cells also can be in one of many cell cycle stages, including GO, Gl, S, G2 or M phase, M phase including the following cell cycle stages: interphase, prophase, prometaphase, metaphase, anaphase, and telophase. Cell components tracked in presently preferable embodiments can include proteins, protein modifications, genetically manipulated proteins, exogenous proteins, enzymatic activities, nucleic acids, lipids, carbohydrates, organic and inorganic ion concentrations, sub-cellular stmctures, organelles, plasma membrane, adhesion complex, ion channels, ion pumps, integral membrane proteins, cell surface receptors, G-protein coupled receptors, tyrosine kinase receptors, nuclear membrane receptors, ECM binding complexes, endocytotic machinery, exocytotic machinery, lysosomes, peroxisomes, vacuoles, mitochondria, Golgi apparatus, cytoskeletal filament network, endoplasmic reticulum, nuclear membrane, proteosome apparatus, chromatin, nucleolus, cytoplasm, cytoplasmic signaling apparatus, microbe specializations and plant specializations.

The following table illustrates some markers and cell components commonly used by embodiments according to the present invention. Other markers can be used in various embodiments without departing from the scope ofthe invention.

Then, in a step 702, one or more samples of the manipulation can be provided to the cells. Manipulations can comprise one or any combination of chemical, biological, mechanical, thermal, electromagnetic, gravitational, nuclear, or temporal factors, for example. For example, manipulations could include exposure to chemical compounds, including compounds of known biological activity such as therapeutics or dmgs, or also compounds of unknown biological activity. Or exposure to biologies that may or may not be used as dmgs such as hormones, growth factors, antibodies, or extracellular matrix components. Or exposure to biologies such as infective materials such as vimses that may be naturally occurring vimses or vimses engineered to express exogenous genes at various levels. Bioengineered vimses are one example of manipulations via gene transfer. Other means of gene transfer are well known in the art and include but are not limited to electroporation, calcium phosphate precipitation, and lipid-based transfection. Manipulations could also include delivery of antisense polynucleotides by similar means as gene transfection. Other genetic manipulations include gene knock-outs or gene mutations. Manipulations also could include cell fusion. Physical manipulations could include exposing cells to shear stress under different rates of fluid flow, exposure of cells to different temperatures, exposure of cells to vacuum or positive pressure, or exposure of cells to sonication. Manipulations could also include applying centrifugal force. Manipulations could also include changes in gravitational force, including sub- gravitation (the preferred embodiment in outer space). Manipulations could include application ofa constant or pulsed electrical current. Manipulations could also include irradiation. Manipulations could also include photobleaching which in some embodiments may include prior addition of a substance that would specifically mark areas to be photobleached by subsequent light exposure. In addition, these types of manipulations may be varied as to time of exposure, or cells could be subjected to multiple manipulations in various combinations and orders of addition. Of course, the type of manipulation used depends upon the application. Then, in a step 704, one or more descriptors of a state in the portions of the cells in the presence of the manipulation can be determined using the images collected on the imaging system. Descriptors can comprise scalar or vector values, representing quantities such as area, perimeter, dimensions, intensity, gray level, aspect ratios, and the like. Other types of descriptors include, but are not limited to, one or any combination of characteristics such as a cell count, an area, a perimeter, a length, a breadth, a fiber length, a fiber breadth, a shape factor, a elliptical form factor, an inner radius, an outer radius, a mean radius, an equivalent radius, an equivalent sphere volume, an equivalent prolate volume, an equivalent oblate volume, an equivalent sphere surface area, an average intensity, a total intensity, and an optical density. These descriptors can be average or standard deviation values, or frequency statistics from the descriptors collected across a population of cells. These descriptors can be further reduced using other methods such as principal component analysis and the like. In some embodiments, the descriptors include features from different cell portions or cell types. That is, a first feature can be from a nuclei and a second feature is from another cell stmcture such as Golgi apparatus, mitochondria, spacing between cell stmctures or cells themselves, as well as many others.

A presently preferable embodiment uses descriptors selected from the following table. Other descriptors can also be used without departing from the scope ofthe invention.

Then, in a step 705, a database of cell information can be provided. Next, in a step 706, a plurality of descriptors can be searched from a database of cell information in order to locate descriptors based upon one ofthe descriptors ofthe manipulation. Then, in a step 708, properties ofthe manipulation are predicted based upon the properties ofthe located descriptors. Properties can comprise toxicity, specificity against a subset of tumors, mechanisms of chemical activity, mechanisms of biological activity, stmcture, adverse biological effects, biological pathways, clinical effects, cellular availability, pharmacological availability, pharmacodynamic properties, clinical uses and indications, pharmacological properties, such as absorption, excretion, distribution, metabolism and the like.

In a particular embodiment, step 706 comprises determining matching descriptors in the database corresponding to a prior administration ofthe manipulation to the descriptors ofthe present administration of the manipulation. In a particular embodiment according to the present invention, combinations of measurements of scalar values can provide predictive information. A database can be provided having one or more "cellular fingerprints" comprised of descriptors of cell- substance interactions of dmgs having known mechanisms of action with cells. Such descriptors can be analyzed, classified, and compared using a plurality of techniques, such as statistical classification and clustering, heuristic classification techniques, a technique of creating "phylogenetic trees" based on various distance measures between descriptors from various dmgs. In this embodiment, numeric values for the descriptors can be used by comparison techniques. A phylogenetic tree can be created that illustrates a statistical significance ofthe similarity between descriptors for the dmgs in the database. Because the dmgs used to build the initial database are of known mechanism, it can be determined whether a particular scalar value in a descriptor is statistically predictive. Finally, a compound descriptor with no known mechanism of action can be queried against the database and be statistically compared and classified among the dmgs in the database that the compound most resembles. In a particular embodiment, relationships between measured morphological properties of images and physiological conditions can be determined. Relationships can include, for example, treatment of different cell lines with chemical compounds, or comparing cells from a patient with control cells, and the like. In a presently preferable embodiment, comparisons can be performed on acquired image features. Some embodiments can comprise statistical and neural network - based approaches to perform comparisons of various features. The foregoing is provided as merely an example, and is not intended to limit the scope ofthe present invention. Other techniques can be included for different types of data. In some embodiments, classification, clustering and other types of predictive data analysis can be performed on features extracted from cell images. In a presently preferable embodiment, statistical procedures for comparisons, classification and clustering are performed on data obtained from imaging cells. Fragments of data preparation and pre-formatting (S language): >tmp . frame < - Generic . Summary

>namesl <- paste ( "Cell . line .5 " , tmp. names, sep=".")

> by . compound. matrix <- as .matrix (arranged. by . compound)

Example ofthe code for principal component analysis (data preparation) using S language: all . data .princomp <- menuPrincomp (data = by . compound. matrix, scores = T, cor = "Correlation", na. action = T, print . short = T, print . importance = T, print . loadings = T, cutoff . loadings = 0.1, plot . screeplot = T, plot . loadings = T, plot.biplot = T, plot .biplot . choices = c(l,2), predict. p = F)

Example of clustering using a divisive hierarchical clustering algorithm: > div.hier .2.manhattan. cluster$call diana (x = tmp . sum. by. comp, diss = F, metric = "manhattan" , stand = T, save.x = T, save. diss = T)

Another embodiment utilizes existing tools for biological sequence similarity searches, classification, and phylogenetic analysis . In a particular embodiment, numbers in a numerical descriptor can be substituted by one or more of nucleic acid or amino acid codes according to a one of several sets of rules. Once converted into a corresponding nucleotide or amino acid sequence representation, the fingerprints can be analyzed and compared using software and algorithms known in the art for genetic and peptide sequence comparisons, such as GCG, a product of Genetics Computer Group, with company headquarters in Madison WI. Select embodiments comprising such approaches enable the use of a broad array of sophisticated algorithms to compare, analyze, and cluster gene and protein sequences. Many programs performing this task are known to those of ordinary skill in the art, such as for example, the PHYLIP (PHYlogeny Interference Package) a package of programs for inferring phylogenies (evolutionary trees) described in (Feldenstein, J. 1996 Methods Enzymol 266:418-427 and Feldenstein, J. 1981 J. Mol. Evol. 17(6):368-376).

Embodiments can perform such analysis based upon factors such as numerical value, statistical properties, relationships with other values, and the like. Further details of a step of manipulation are noted more particular below. Fig. 7B illustrates a representative block flow diagram of simplified process steps for determining one or more descriptors of a state in the portions ofthe cells in the presence ofthe manipulation of step 704 of Fig. 7 A in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. In a step 712, an image of a cell portion is obtained. In some embodiments, the cell portion is visualized with a fluorescently labeled marker that is specific for the portion or portions of interest. A cell portion can include, for example, one or more ofthe following: nuclei, Golgi apparatus, and other features. The cell portion may vary in select embodiments according to the invention. Then, in a step 714, a digitized representation ofthe image obtained in step 712 is determined. In some embodiments, steps 714 and step 712 can comprise a single step. These embodiments use a digital imaging means such as a digital camera, to obtain a digital image ofthe target directly. Next, in a step 716, the digital representation ofthe image is processed to obtain image features. Image features can include such quantities as area, perimeter, dimensions, intensity, aspect ratios, and the like. Then, in a step 718 descriptors can be determined from the image features. Descriptors can comprise scalar or vector quantities and can comprise the image features themselves, as well as composed features, such as shape factor derived by a relationship 4π * area / perimeter, and the like. Descriptors can also comprise statistical quantities relating to feature characteristics across a population of cells, such as a standard deviation, and average, and the like. In a preferred embodiment, cells can be placed onto a microscope, such as a Zeiss microscope, or its equivalent as known in the art. A starting point, named Site A01 , is identified to the microscope. A plurality of exposure parameters can be optimized for automated image collection and analysis. The microscope can automatically move to a new well, automatically focus, collect one or more images, at one or more wavelengths, move to a next well, and repeat this process for all designated wells in a multiple well plate and for multiple plates. A file having a size and an intensity distribution measurement for each color and rank for each well can then be created for the images acquired. Based on this information, a user or a computer can revisit sites of interest to collect more data, if desired, or to verify automated analysis. In a presently preferred embodiment, image automatic focus and acquisition can be done using computer software controlling the internal Z-motor of the microscope. Images are taken using a lOx, 20x, or 40x air long working distance objectives. Sometimes multiple images are collected per well. Image exposure times can be optimized for each fluorescent marker and cell line. The same exposure time can be used for each cell line and fluorescent marker to acquire data.

Fig. 7C illustrates a representative block flow diagram of simplified process steps for obtaining images of cell portions of step 712 of Fig. 7B in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The method is generally outlined by the steps below:

(1). In a step 720, a sample is provided to the imaging device. Samples can be provided in 96 well plates and the like. The sample may be loaded into a microscope, such as a Zeiss microscope or equivalent. (2). In a step 722, a set of optical filters is selected to shine light ofthe appropriate wavelength to illuminate the first sample, which may be contained in a first well designated A01. (3). In a step 724, an automatic focusing procedure is performed for the site. In a particular embodiment, the internal z-motor of the microscope which is attached to the objective nosepiece is used for automatic focusing of the microscope. In an alternative embodiments, the plate holding the samples is moved to perform automatic focusing ofthe microscope, or focusing can be performed by moving optical components attached to the microscope and the like.

(4). In a step 726, images are collected for the site. Images can be collected for every color at every site. Present embodiments can provide images for up to four colors. However, embodiments are contemplated that can provide more colors by using either a monochromator coupled with excitation filters which are on a filter wheel, or by digitally separating overlapping fluorophores. Those knowledgeable in the field will know that given calibration images of single fluorophores, a look-up table can be devised which will allow for the digital removal of fluorescence bleed-though of fluorescence which may occur in optical channels other than the one for which that filter has been optimized in instances of using more than one fluorophore at once. Cell growth and density information is also collected. Cell density is determined by what percentage ofthe area being imaged is inhabited by cells. In some embodiments, imaging can be facilitated using one or more biosensors, molecules such as non-proteins, i.e., lipids and the like, that are luminscently tagged. However, some embodiments can also use fluorescence polarization and the like. Fluorescence polarization is a homogeneous fluorescence technology where the excited state ofthe molecule lasts much longer than in normal fluorescence, taking seconds to minutes to reach equilibrium, obliterating the need to wash away fluorescence markers that are not specifically bound to a marker. Further, embodiments can detect differences in spectral shifts of luminescent markers. Some fluorescence markers, such as Nile Red sold by Molecular Probes of Eugene, OR, will change its emission peak wavelength depending on its environment. One can detect these changes by monitoring the level of fluorescence at both wavelengths and reading out at ratio of the two. (5). In a step 728, a determination is made whether more fields of view need to be taken for a particular color. If this is so, then processing continues at step 726 at a new site. Otherwise, processing continues with a decisional step 730. Images can now be taken by repeating step 726. In a preferred embodiment 4 to 9 images are collected at each site.

(5). In a step 730, a determination is made whether more optical configurations need to be taken in order to obtain images for all differently-marked cell portions the sample. If this is so, then in a step 732 a new optical configuration is determined. Images for the new optical configuration can now be taken by repeating steps 726 and 728.

(6). In a decisional step 734, after all optical configurations and images for fields of view in a sample have been obtained, a determination is made whether any further samples remain to be analyzed. If so, a new sample is brought into view and processing continues with step 720. Otherwise, image processing is complete. In a presently preferable embodiment, image data can be stored on a CD ROM using a CD ROM burner, such as CRW4416 made by Yamaha of Japan. However, other mass storage media can also be used. Fig. 7D illustrates a representative block flow diagram of simplified process steps for processing digitized representations of step 716 of Fig. 7B in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The method is generally outlined by the steps below:

(1). In a step 740, a digitized image input is preprocessed . Preprocessing might include, but is not limited to, such operations as background subtraction, thresholding, smoothing, adoptive filtering, edge enhancements, contrast enhancements, histogram equalization. A particular combination of preprocessing steps can be applied to images in successive steps or in parallel to copies ofthe image.

A simplified example of a smoothing and background subtraction procedure in a MatLab language is presented in computer code below:

function Isubtracted = cmBac grSubtrl (I , k) % cmBackgrSubtrl (I , k) - simple flat background (=modal*k) subtraction % Y = cmBackgrSubtrl (I , k) - image Y is generated by % subtraction (with saturation) of modal pixel value of I multiplied by k % DEFAULT - k=l

Ό if (nargin == 1) k=l; end if (sιze(k)~=l) error (' cmBackgrSubtrl : parameter k should be a number. Exiting... ' ) ; end

%modpιxnum = floor ( size ( I ( : ) , 1 ) /2 ) ; %sortedval = sort ( double (I ( : ) ) ⁾ ; %modpιxel = sortedval (modpixnum) ; modpixel = median (double (I (:))) ; bg = k*modpιxel ;

Isubtracted = mmsubm ( umt8(I), umt8 (round (ones ( size (I) ) *k*modpιxel )) );

An example of a procedure for thresholding m computer code (MatLab) is presented below: function thresh = GetThreshByPenml ( I , ) % GetThreshByPenml (I) Finds optimal thresholding value for image I

% N = GetThreshByPenml (I) Finds thresholding value N for image I

% N = GetThreshByPenml (I , M) - tests threshold values up to M

% DEFAULT M = maximum pixel value m I

% note that GetThreshByArea is significantly faster

% finds a threshold value that causes the maximal change m the % total perimeter of the objects (Russ ????) % see Matlab_Auto_thresholdl_l-23-99.doc for more details % Note: works somewhat better on SMOOTH images (i.e. medfιlt2(I, [3 3]) two times

if (nargm == 0) error (strcat ( mfilename, at least one parameter required ' ) ) ; elseif (nargm == 1)

M = double (max (I (:))); %test thresholds up to maximum pixel value m I elseif (nargm > 2) error (strcat (mfilename, too many parameters ' ) ) ; end

if (sιze(M)>l) error (strcat (mfilename, argument M should be a number ' ) ) ; end

Mmval = double ( mm ( I ( : ) ) ) ; step = 1 ;

%generate vertical vector perims with total perimeters of objects at different %threshold values for ι=Mmval : step : M bwl = ιm2bw(I, ι/255) ; prl = bwpeπm (bwl ) ; pr = sum (prl ( : ) ) ; if (exist ( 'perims ' , 'var') == 0) %peπms is yet undefined perims = pr; else perims = cat ( l , perims , pr) ; end end

% vector prdi f f s contains di f f erences between success ive perimeters prdi f f s = dif f (perims ) ; mmdecrease = mm (prdi f f s ) ; mmvalues = f ind (prdi f f s == mmdecrease ) ; ιndex_of_mmde crease = mmvalues ( 1 ) ; thresh = mdex_of_mmdecrease + 1 ;

% = ------ -- = = = = = ------.= end GetThreshl = = = = = = = = = = = = = = = = = =

Thresholding provides a specific intensity, such that pixels darker than the threshold are deemed black, and pixels lighter than the threshold are considered white The thresholded image can be processed using binary image processing techniques in order to extract regions

(2) In a step 742+744, the digitized image input is subjected to object identification This can be accomplished by a vaπety of procedures, for example by thresholding or edge detection and subsequent morphological opening and closing Edge detection can be accomplished by means of gradient-based or zero-crossing methods, such as Sobel, Canny, Laplassian, Perwitt, and other methods

An example of object identification procedure based on Canny edge detection (in MatLab language) is presented below

function Imask = cmMaskDNAl (I) ;

% cmMaskDNAl - generates binary mask for cell nuclei through edge detection % Imask = cmMaskDNAl (I )

% PARAMETERS

% I - intensity image (grayscale)

% OUTPUT

% Imask - BW image with objects from I g, o

% For more details see Notebook Matlab_DNA_maskιngl_l-22- 99.doc

% Uses SDC Morphology Toolbox VO .7

if (nargm -= 1) error (' rong number of input parameters'), end if (nargout -= 1) error ('Wrong number of output parameters: one output argument should be provided') ; end

Imask = edge (I, 'canny');

Imask = mmdil (Imask, mmsecross (1) ) ;

Imask = mmero ( mmclohole (Imask, mmsecross (1) )) ;

Imask = mmedgeoff (Imask, mmsecross (1) ) ;

% note that mmedgeoff this command removed FILLED OBJECTS but not touching OUTLINES.

% these outlines can be removed by filtering:

Imask = medfιlt2 (Imask, [5 5] ) ;

=end cmMaskDNAl

However, embodiments can also use other techniques, such as Fast Fouπer Transforms (FFT) and the like as known in the art without departing from the scope of the present invention (3) In a step 746, a plurality of region features can be determined For example, m a representative embodiment, image features can include such quantities as area, peπmeter, dimensions, intensity, aspect ratios, and the like Features not directly related to individual objects are also being extracted An example of a procedure for extraction of some of the features (MatLab language) is presented below

function OData = cmGetObj ectsData (I , Ilabel) % cmGetObj ectsData returns array measurements of objects m image "I" masked by "Ilabel"

% EV 2-3-99; 2-10-99

% OData = cmGetObj ectsData (I , Ilabel) returns an array of morphological and intensity measurements % taken from a grayscale image "I". Objects are identified on a mask image Ilabel, usually

% created by bwlabelO

% OUTPUT:

% Each row m the output array OData represents individual object

% columns contain the following measurements:

%

% 1 - Index ("number" of an object) ; 8

Solidity; % 2 - X coordinate of the center of mass; 9 - Extent;

% 3 - Y coordinate -"- ; 10 - Total

Intensity;

% 4 - Total Area (m pixels); 11 - Avg .

Intensity; % 5 - Ratio of Maj orAxis/MmorAxis ; 12 - Median

Intensity;

% 6 - Eccentricity; 13 - Intensity of

20% bright pixel

% 7 - EquivDiameter ; 14 - Intensity of 80% bright pixel o, o

% For details on morphological parameters see information on MatLab lmfeature ( ) ; % Intensity parameters are either obvious or are documented m comments m this file.

if (nargm ~= 2) error ('function requires exactly 2 parameters'); end if (nargout ~= 1) error ('function has 1 output argument (array X by 14) ') ; end

% finished checking arguments

% first collect morphological parameters m a structure array:

ImStats = lmfeature (Ilabel , 'Area', 'Centroid', ¹ MajorAxisLength ' , ...

'MmorAxisLength' , 'Eccentricity', ' EquivDiameter ' ,

'Solidity', 'Extent', 8 ) ;

% now convert it into array (matrix) while collecting intensity data for each object:

%preallocate output array: numobjects = size (ImStats, 1) ; OData = zeros (numobjects , 14);

%now convert ImStats into array and add intensity data to it for k=l : numobj ects

OData (k, 1) = k;

OData(k, 2) = ImStats (k) . Centroid ( 1 ) ; OData(k, 3) = ImStats (k) . Centroid (2 ) ; OData(k, 4) = ImStats (k) .Area ; OData(k, 5) = ( ImStats (k) .Maj orAxisLength) / (ImStats (k) .MmorAxisLength) ;

OData (k, 6) = ImStats (k) .Eccentricity ;

OData(k, 7) = ImStats (k) . EquivDiameter ; OData(k, 8) = ImStats (k) . Solidity ;

OData(k, 9) = ImStats (k) .Extent ;

% now collect and assign intensity parameters from image I

object_pιxels = fmd( Ilabel == k) ; ob ect_area = size (obj ect_pιxels , 1) ; %same as total number of pixels m the object object_mtensιtιes = double (I (ob ect_pιxels) ) ; % need to convert to double to do math sorted_mtensιtιes = sort (object_mtensιtιes) ; % will need to get median, 20% and 80% pixels total_mtensιty = sum (object__mtensιtιes , 1) ; avg_mtensιty = total_mtensιty / object_area; medιan_mtensιty = sorted_mtensιtιes ( f loor ( obj ect_area/2 ) + 1 ) ; pιx20 = sorted_mtensιtιes ( f loor (obj ect_area*0.2 ) +1 ) ; %bπghtest pixel among dimmest 20% pιx80 = sorted_mtensιtιes ( f loor (obj ect_area*0.8) +1 ) ;

OData (k, 10) = total_mtensιty ;

OData (k, 11) = avg_mtensιty ;

OData (k, 12) = medιan_mtensιty ; OData (k, 13) = pιx20; %bπghtest pixel among dimmest

20%

OData (k, 14) = pιx80; %dιmmest pixel among brightest 20% end %for % =--====-_=======------== end function cmGetObj ectsData ()===------====--=----===------===------===----

(4). In a step 748, quantitative descriptors, characterizing cell state are calculated based on the feature measurements extracted at step 746. For example, histogram distribution of intensities of cell nuclei provides information about the population cell cycle stages.

In a particular embodiment according to the present invention, data analysis techniques for describing the fluorescence patterns of cell portions in multiple cell lines in the presence and absence of compounds are provided. Automated image analysis techniques can include determining one or more regions from around nuclei, individual cells, organelles, and the like, called "objects" using a thresholding function. Objects that reside on the edge of an image can be included or excluded in various embodiments. An average population information about an object can be determined and recorded into a database, which can comprise a database text file or Excel spreadsheet, for example. However, embodiments can use any recording means without departing from the scope ofthe present invention. Values measured can be compared to the visual image. One or more types of numerical descriptors can be generated from the values. For example, descriptors such as a number of objects, an average, a standard deviation of objects, a histogram (number or percentage of objects per bin, average, standard deviation), and the like can be determined.

In a particular embodiment according to the present invention, data can be analyzed using morphometric values derived from any of a plurality of techniques commonly known in the art. For example, a software package called MetaMorph Imaging System, provided by Universal Imaging Corporation, a company with headquarters in West Chester, PA and NIH Image, provided by Scion Corporation, a company with headquarters in Frederick, Maryland. Fluorescent images can be described by numerical values, such as for example, an area, a fluorescence intensity, a population count, a radial dispersion, a perimeter, a length, and the like. Further, other values can be derived from such measurements. For example, a shape factor can be derived according to a relationship 4π * area / perimeter. Other values can be used in various embodiments according to the present invention. Such values can be analyzed as average values and frequency distributions from a population of individual cells.

In a particular embodiment according to the present invention, techniques for the automatic identification of mitotic cells are provided. Image analysis techniques employing techniques such as multidimensional representations, frequency-based representations, multidimensional cluster analysis techniques and the like can be included in various embodiments without departing from the scope of the present invention. Techniques for performing such analyses are known in the art and include those embodied in MatLab software, produced by Math Works, a company with headquarters in Natick, MA.

Scalar values providing efficacious descriptors of cell images can be identified using the techniques of the present invention to perform predictive analysis of drug behavior. In a presently preferred embodiment, a plurality of heterogeneous scalar values can be combined to provide descriptors for each manipulation. By applying predictive analysis routines to the collections of these descriptors, predictive information about any number of manipulations and cell interactions can be extracted.

Fig. 7E illustrates a representative block flow diagram of simplified process steps for analyzing image feature values to obtain descriptors of cell state of step 718 of Fig. 7B in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Fig. 7E illustrates an input data of descriptors of known manipulations 319. A step 320 of reformatting and transforming data 319 to formats suitable for analysis is performed. Additionally, a "cleaning" process can eliminate outlying data points and the like in the data. Then, in a step 322, a decision is made whether to continue with step 324 or with step 326 based upon determining a particular type of analysis appropriate for the present application or particular type of prediction. If decisional step 322 determines processing should continue with step 324, then, in that step, an error estimate using a set of test descriptors is performed to estimate the quality of a prediction and processing continues with step 320. Once an optimal prediction is achieved, processing continues with step 326. In step 326, optimal transformation parameters and prediction methods are selected for use in steps 328 and 330 which analyze data about an unknown manipulation. In a step 328, a solution is generated based upon any of techniques including training a neural network, solving a mathematical equation, applying decision tree mles and/or the like. In a step 330, an input data set of unknown descriptors 318 is reformatted and transformed based upon the optimal transformation parameters selected in step 326 using the transformation procedures in steps 320, 322 and 324. In a step 332, predictions techniques are applied to the reformatted manipulations from step 330 and the solution generated in step 328 and a plurality of properties of known manipulations 317 (e.g., therapeutic properties, and the like) in order to determine a prediction of properties of unknown manipulation 316.

Fig. 7F illustrates a representative block flow diagram of simplified process steps for a method of mapping a manipulation of cells to a physiological characteristic in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The method is generally outlined by the steps below:

(1) In a step 750, a plurality of cells, e.g., dead, live, cell fractions or mixtures of cells are provided.

(2) Then, in a step 752, the plurality of cells is manipulated, where manipulation occurs using a source(s) from one or a combination selected from an electromagnetic, electrical, chemical, thermal, gravitational , nuclear , temporal , or a biological source.

(3) Next, in a step 754, a feature value is captured from the plurality of cells. The feature value can include one or any combination of characteristics such as cell count, area, perimeter, length, breadth, fiber length, fiber breadth, shape factor, elliptical form factor, inner radius, outer radius, mean radius, equivalent radius, equivalent sphere volume, equivalent prolate volume, equivalent oblate volume, equivalent sphere surface area, average intensity, total intensity, and optical density. This list is not meant to be limiting. (4) Then, in a step 756, a degree of presence of one or more feature values is assigned for each manipulation.

(5) In a step 758, the feature values from the plurality of cells are stored in memory locations. From the memory locations the values can be used for statistical analyses to produce predictive information about the relatedness ofthe descriptors of the manipulations to one another. This information is used to infer properties of the manipulations.

Fig. 7G illustrates a representative block flow diagram of a simplified process steps for a method for populating a database with manipulated biological cell information in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. The method is generally outlined by the steps below: (1) In a step 760, a plurality of cells in various stages of the cell cycle,

A montage image that was used as a source to generate data in Appendix A is presented in Fig. 12., such as for example, the stages of interphase, prophase, metaphase, anaphase, and telophase are provided.

(2) Then, in a step 762, each of the cells in the various stages of mitotic development is manipulated.

(3) Next, in a step 764, an image ofthe plurality of manipulated cells is captured using image acquisition techniques in order to provide a morphometric characteristic of each ofthe manipulated cells.

(4) As a preferable option, in a step 766, an image database may be populated with the image of the plurality of manipulated cells.

(5) Following step 764 or optional step 766, a morphological value is calculated from the image in a step 768.

(6) In a step 770, the database is populated with the morphological value. Fig. 7H illustrates a representative block flow diagram of simplified process steps for a method for populating a database with manipulated biological information, e.g., image acquisition parameters, image feature summary information, and well experimental parameters in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope of the claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Fig. 7H illustrates a step 780 in which cells are placed into site on a plate and a manipulation is applied. Then, in a step 781 an image is taken of the cells. In step 782, the image is transferred to an image archive database. Then, in a step 783, well experimental parameters are entered into the database 787. Well experimental parameters can include cell type, manipulation and the like. In a step 784, image acquisition parameters are transferred to database 787. Image acquisition parameters can include file name, fluorophores and the like. In a step 785, the image acquired in step 781 is analyzed. Then, in step 786, an image feature summary from the analysis step 785 is transferred to database 787.

In step 788, a lookup table for all analyses is provided to database 787. The lookup table provides information about the analyses. In a step 789, a query of database 787 for process data is performed. The results are reformatted. Then in a step 790, the database 787 is queried. Next, in a step 791, features of the manipulations stored in the database are combined and reduced. Next, in a step 793, reduced features of step 791 can be compared. In a step 792, the results of step 793 are recorded in database 787. Then, in a step 794, a report of predictions based on comparisons performed in step 793 is generated. Fig. 71 illustrates a representative block flow diagram of simplified process steps for acquiring images of manipulated biological information, e.g., cells, cell tissues, and cell substituents in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Fig. 71 illustrates a step 770 in which a user sets up an image analysis procedure. Then, in a step 772, an image is read into image analysis software. Next, in a step 774, patterns and objects are identified in the image using one or more algorithms. Next, in a step 776, sets of features are extracted from the image. Then, in a step 778, feature information, descriptor values and the like are exported to the database, such as database 787 of Fig. 7H, for recording. Next, in a decisional step 779, a determination is made whether any more images should be taken. If this is so, processing continues with step 772. Otherwise, image acquisition processing is completed.

Fig. 7J illustrates a representative block flow diagram of simplified process steps for populating, acquiring and analyzing images of manipulated biological information in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Fig. 7J illustrates a step 300 of placing a plate onto an imaging stage and reading a bar code. Then, in a step 301 an autofocus procedure is performed. Next, in a step 302, a first optical filter configuration is selected and an image is collected. Then, in a decisional step 303, a determination is made whether more than one image per optical configuration can be taken. If so, then, in a step 304, a new position within the well is targeted and another image is collected. Then, in a decisional step 305, a determination is made whether any more images need to be collected. If this is so, step 304 is repeated until all images for a particular well have been collected. After one or more images are collected for the well, in a step 306, the stage is returned to a starting position within the well, and a montage is created from collected images. The results are named with a unique file name and stored.

In a decisional step 307, a determination is made whether any more optical channels in the well can be imaged. If this is so, then in a step 308 the next optical filter configuration is selected and an image is collected. Processing then continues with decisional step 303, as described above. Otherwise, if no further optical channels in the well can be imaged, then in a decisional step 309 a determination is made whether any wells remain to be imaged. If not all wells have been imaged, then in a step 310, the stage moves to the next well and processing continues with step 301, as described above. Otherwise, if all wells on the plate have been imaged, then in a decisional step 311, a determination is made whether any more plates can be processed. If this is so, then processing continues with step 300 as described above. Otherwise, in a step 312, the information is stored to a CD or other storage device as a backup.

Fig. 7K illustrates a representative block flow diagram of simplified process steps compound based upon information about effects of one or more known compounds on a cell population in a particular embodiment according to the present invention. This diagram is merely an illustration and should not limit the scope ofthe claims herein. One of ordinary skill in the art would recognize other variations, modifications, and alternatives. Fig. 7K illustrates a step 340 of populating a database with descriptors for known compounds. Such descriptors can be determined from imaging the cell population. However, in some embodiments, descriptors can be derived by measurements and combinations of measurements and the like. Then, in a step 342, descriptors for the unknown compound are determined from imaging a second cell population. The second cell population has been treated with the unknown compound. Then, in a step 344, a relationship between the descriptors determined from the unknown compound with the descriptors determined from the known compounds can be determined. Finally, in a step 346, an inference can be made about the unknown compound based upon the descriptors of the known compounds from the relationship determined in step 344.

Accordingly, the present invention provides a novel database design. In a particular embodiment according to the present invention, a method for providing a database comprises measurement of a potentially large number of features of one or more sub-cellular morphometric markers. Markers can be from any of a large variety of normal and transformed cell lines from sources such as for example, human beings, fungi, or other species. The markers can be chosen to cover many areas of cell biology, such as, for example markers comprising the cytoskeleton of a cell. The cytoskeleton is one of a plurality of components that determine a cell's architecture, or "cytoarchitecture". A cytoarchitecture comprises stmctures that can mediate most cellular processes, such as cell growth and division, for example. Because the cytoskeleton is a dynamic stmcture, it provides a constant indication of the processes occurring within the cell. The cytoarchitecture of a cell can be quantified to produce a one or more scalar values corresponding to many possible cellular markers, such as cytoskeleton, organelles, signaling molecules, adhesion molecules and the like. Such quantification can be performed in the presence and absence of dmgs, peptides, proteins, anti-sense oligonucleotides, antibodies, genetic alterations and the like. Scalar values obtained from such quantification can provide information about the shape and metabolic state of the cell. In a presently preferred embodiment, scalar values can comprise morphometric, frequency, multi-dimensional parameters and the like, extracted from one or more fluorescence images taken from a number of cellular markers from a population of cells. Two or more such scalar values extracted from a plurality of cell lines and markers grown in the same condition together comprise a unique "fingerprint" or descriptor that can be incorporated into a database. Such cellular descriptors will change in the presence of dmgs, peptides, proteins, antisense oligonucleotides, antibodies or genetic alterations. Such changes can be sufficiently unique to permit a correlation to be drawn between similar descriptors. Such correlations can predict similar properties or characteristics with regard to mechanism of action, toxicity, animal model effectiveness, clinical trial effectiveness, patient responses and the like. In a presently preferred embodiment, a database can be built from a plurality of such descriptors from different cell lines, cellular markers, and compounds having known mechanisms of action (or stmcture, or gene response, or toxicity).

The present invention also provides database and descriptor comparisons according to other embodiments. In a particular embodiment according to the present invention, measurement of scalar values or features can provide predictive information. A database can be provided having one or more "cellular fingerprints" comprised of descriptors of cell substance interactions of drugs having known mechanisms of action with cells. Such descriptors can be compared using a plurality of techniques, such as a technique of creating "phylogenetic trees" of a statistical similarity between the descriptors from various dmgs. In a present embodiment, scalar, numeric values can be converted into a nucleotide or amino acid letter. Once converted into a corresponding nucleotide representation, the descriptors can be analyzed and compared using software and algorithms known in the art for genetic and peptide sequence comparisons, such as GCG, a product of Genetics Computer Group, with company headquarters in Madison WI. In an alternative embodiment, numeric values for the fingerprints can be used by comparison techniques. A phylogenetic tree can be created that illustrates a statistical significance ofthe similarity between descriptors for the dmgs in the database. Because the dmgs used to build the initial database are of known mechanism, it can be determined whether a particular scalar value in a descriptor is statistically predictive. Finally, a compound fingerprint with no known mechanism of action can be queried against the database and be statistically compared and classified among the dmgs in the database that the compound most resembles.

In a particular embodiment, relationships between measured morphometric properties and features of images and physiological conditions can be determined. Relationships can include, for example, treatment of different cell lines with chemical compounds, or comparing cells from a patient with control cells, and the like. In a presently preferable embodiment, a clustering can be performed on acquired image descriptors. Some embodiments can comprise statistical and neural network - based approaches to perform clustering and comparisons of various descriptors. The foregoing is provided as merely an example, and is not intended to limit the scope ofthe present invention. Other techniques can be included for different types of data. In some embodiments, clustering and comparing can be performed on features extracted from cell images. In a presently preferable embodiment, procedures for comparisons and phylogenetic analysis of biological sequences can be applied to data obtained from imaging cells.

Select embodiments comprising such approaches enable the use of a broad aπay of sophisticated algorithms to compare, analyze, and cluster gene and protein sequences. Many programs performing this task are known to those of ordinary skill in the art, such as for example, the program Phylip, available at http://evolution.genetics.washington.edu/phylip.html. and other packages listed at http://evolution.genetics.washington.edu/phylip/software.html . However, select embodiments according to the present invention can comprise a technique of statistical classification, statistical clustering, distance based clustering, linear and non-linear regression analysis, self-organizing networks, and mle-based classification.

Embodiments can perform such analysis based upon factors such as numerical value, statistical properties, relationships with other values, and the like. In a particular embodiment, numbers in a numerical descriptor can be substituted by one or more of nucleic acid or amino acid codes. Resulting "pseudo-sequences" can be subjected to analysis by a sequence comparison and clustering program.

Other types of databases can also be provided according to other embodiments. The database includes details about the properties of a plurality of standard drugs. When the descriptor of a test compound is compared to the database, predictions about the properties ofthe test compound can be made using any known property of the other compounds in the database. For example, properties about a compound in the database could include stmcture, mechanism of action, clinical side effects, toxicity, specificity, gene expression, affinity, pharmacokinetics, and the like. The descriptor of a compound of unknown stmcture from a natural products library could be compared to the descriptors of compounds with known stmcture and the stmcture could be deduced from such a comparison. Similarly, such information could lead to better approaches to dmg discovery research including target validation and compound analogizing, as well as pre-clinical animal modeling, clinical trial design, side effects, dose escalation, patient population and the like.

According to the present invention, databases can be integrated with and complementary to existing genomic databases. Differential genomic expression strategies can be used for dmg discovery using database technology. In one particular embodiment, cell data and cellular response data can be associated with a genetic expression profile assay to form a single assay. Live cells expressing fluorescence markers can be treated with a dmg, imaged and analyzed for morphometry; and then analyzed for mRNA for expression. Such embodiments can provide rapid development of tools to link cellular behavior with functional genomics.

Database methods according to the present invention can be used to predict gene function and to assist in target validation. Databases that include genetic diversity, i.e., having cellular descriptors from cells of differing genetic backgrounds (tumor, tissue specific, and gene knock out cell lines), can provide the capability to compare cells of unknown genetic background to those in the database. Similarly, the descriptor of an unknown cellular portion in the presence of multiple dmgs can be queried against the descriptors ofthe known markers in the database. For example, if an unknown gene is tagged with Green Fluorescent Protein (GFP), the database may be used to identify the cellular portions for which that unknown gene encodes. According to the present invention, target validation and specialized cell-based assay screening can be performed using database systems and methods to serve as a universal high-throughput cell-based assay that can evaluate the molecular mechanism of dmg action. As new genes are isolated and identified, a large collection of available gene-based knowledge is becoming available. From this large collection of new genes, potential protein targets can be identified using the genomic tools of sequence analysis and expression profiling. However, unless a gene mutation is tightly linked to a disease state, further validation of individual targets is a time consuming process, becoming a bottleneck in dmg discovery. Furthermore, robotics and miniaturization are making "High Throughput Screening (HTS)" the industry standard, substantially reducing the time and cost of running a target-based biochemical assay. Therefore, it is now possible to routinely screen large libraries and use a resulting "hit" to validate the target. In such approaches, a specialized cell- based assay would be developed to test hits for each target. Since this often involves the creation of cell lines expressing new markers, this stage may also become a bottleneck that cannot keep pace with HTS. In addition, these cell-based assays may not be amenable to high-throughput screening, making it difficult to test the increasing number of analogs arising from combinatorial chemistry. In a particular embodiment according to the invention, a rapid characterization of large compound libraries for potential use as pharmaceutical products can be provided by predicting properties of compounds that relate to the compounds' potential as bioactive dmgs. In many dmg discovery situations, virtually millions of compounds can be passed through a HTS assay against a small number of validated targets. These assays produce hundreds to thousands of potential hits.

These hits can then be subsequently screened by a pipeline of secondary and tertiary screens to further characterize their specificity, often time completely missing nonspecific interactions with other proteins. Techniques according to the present invention can provide a replacement to such screening operations by providing information about cellular accessibility and mechanism of action for the hits coming from a HTS system. Furthermore, it can replace the biochemical HTS assay and allow rapid and accurate identification of attractive compounds from large libraries without an intervening biochemical assay. The cell information can be predictive of whether to continue into an animal model for each compound, and which animal model to pursue.

The principles ofthe present specifically contemplate a wide variety of research methodologies, or usage scenarios, implementing these principles. The following discussion of three such scenarios is by way of illustration and not limitation. Study ofthe principles enumerated herein will render evident to those skilled in the art certain additional methodologies or usage scenarios enabled by the teachings hereof. The present invention specifically contemplates all such modifications. The following description presents some specific embodiments and scenarios that represent a broader use of cellular phenotypic data and characterizations to deduce mechanisms of action and other features of cellular responses to various stimuli. Such procedures generally involve producing a quantitative cellular phenotype based upon two or more cellular attributes and then comparing that phenotype to phenotypes previously stored and indexed. Such procedures make use of databases or other repositories of biological information. The invention is not limited to the specific embodiments described here.

Considering first the procedure 2000 depicted in Figure 20, a compound has been identified as having a particular cellular activity. See 2004. For example, a compound may be found to inhibit the growth of certain cancer cell in vitro by a specific and desired mechanism of action. This may be a particular company's "gold standard."

Next, the compound is analyzed at 2006 in terms of its effect on one or more cell lines. More specifically, the compound is linked, virtually, to a particular phenotype. Two or more values or measures of cellular attributes characterize that phenotype. These attributes are quantified in the context of specific cellular markers.

In one example, the cellular marker is an organelle such as a nucleus or Golgi apparatus. Measured attributes useful for characterizing an associated phenotype include geometric parameters (e.g., size, shape, and/or location ofthe organelle) and composition (e.g., concentration of particular biomolecules within the organelle).

The phenotype may be characterized by administering the compound of interest to various cell lines and in various concentrations. In each example within this matrix, the attributes of interest are measured. Ultimately, certain phenotypic features (combinations of attribute values) are associated with the compound of interest. These features provide a template for the phenotype.

Next, using the phenotype as identified at 2006, the process identifies other compounds providing similar features. The goal here is to present a list of compounds having a mechanism of action similar to that of the compound that started the process. This allows researchers to identify a mechanism of action, if not already known, for their compound and to draw conclusions based upon their compound's link to other known compounds (which may not be chemically/structurally similar to the compound of interest).

Identifying similar compounds based upon phenotype can take many paths. Most will involve some mathematical basis. For example, the phenotype defined at 2006 can be represented as a fingerprint or vector comprised of multiple scalar values of cellular attributes (as described above). The phenotype representation can then be compared against known phenotypes characterized by the same format (e.g., they are all characterized as vectors having the same attribute set, but with different values of the attributes). The comparison may be as simple as a Euclidean distance or more sophisticated as a neural network or multivariate statistical correlation. The known compounds and associated phenotypes may be stored as database records or other data stmctures that can be queried or otherwise accessed as part ofthe identification procedure. The compounds may also be associated with other relevant data such as clinical toxicity, cellular toxicity, hypersensitivity, mechanism of action, etc. (when available). Compounds found to be sufficiently similar to the starting compound are returned for consideration by researchers. A data processing system may rank such compounds based on degree of similarity to the starting compound. In some cases, the system may even provide similarity scores associated with the listed compounds. Often researchers wish to determine whether their particular compound has clinical or biochemical effects beyond those that they are already aware of. In a typical scenario, the compound of interest was selected based upon its strong binding a target or its stimulation or inhibition of cell growth in a particular cell line. The process associated with 2010 has likely identified the compound of interest as having a particular mechanism of action based on phenotypic similarity to other compounds having a similar mechanism of action. However, within the region of biochemical space, there may be subspaces (characterized by subphenotypes) that correspond to separate properties. For example, within the phenotypic space associated with one mechanism of action, there may be subspaces associated with clinical toxicity, cellular toxicity (likely overlapping the clinical toxicity space), and little or no toxicity. Obviously, a researcher would like to know whether her compound is likely to be toxic.

Thus, the process 2000 may include characterizing the compound of interest in terms of its distance from (i.e., similarity to) specific phenotypes having known characteristics. In a typical example, the known characteristic is toxicity. This feature allows the researcher to quantify her compound in terms of mechanism of action AND toxicity (or in terms of two or more other relevant properties associated with phenotype). To allow simple ranking or characterization, compounds of interest may be scored according to a simple or weighted Boolean expression.

A second scenario of interest is depicted in Figure 21. This scenario again defines a phenotype in terms of a quantifiable vector or other measure. However, rather than using a compound of interest to generate the phenotype, some other cellular stimulus is used to generate the phenotype.

As shown, a process 2100 begins with receipt of cells of interest. See 2104. In many situations, the cells are produced by a genetic or epigenetic process that affects the expression level or activity of a particular protein. More generally, any cellular stimulus (e.g., radiation level and type, gravity level, magnetic field, acoustic perturbations, etc.) can be used to generate the cell line of interest. Importantly, this stimulus affects the phenotype and can be correlated therewith.

In the context of dmg discovery, a gene encoding for a particular target can be genetically knocked out, underexpressed, overexpressed, expressed in a non- native state, etc. This may be accomplished via standard procedures involving genomic modification, translation or transcription apparatus modification (e.g., use of antisense nucleic acids), blocking target activity (using antibodies to a receptor site for example), and the like. These processes will generally affect the phenotype in some quantifiable way. Importantly, they clearly and unambiguously define a cellular phenotype associated with altering the activity ofthe target protein.

At 2106, the process involves measuring one or more cellular features from the cell line of interest to define/quantify the phenotype. This may be accomplished as described above with reference to 2006. Next, at 2108, the cellular phenotype generated in this manner is used to identify and rank a set of compounds associated with the phenotype. This operation may proceed in the manner of operations 2008 and/or 2010 from Figure 20.

Finally, at 2110, the process clusters the compounds returned at 2108 by a mechanism of action. The operation 2106 has tightly bound a mechanism of action to a phenotype. Various compounds characterized and stored in a system database may be tentatively assigned a mechanism of action or may have no suggested mechanism of action. By matching their virtual phenotype to the phenotype generated at 2106, one can create or strengthen an association between the compounds and mechanism of action relevant to the stimulus at 2104. Considering now Figure 22, a third scenario is depicted. This scenario again involves using a virtual phenotype to glean information relevant to a mechanism of action or other cellular activity. In this case, assay data from a group of compounds (e.g., a primary or focused library) is used to elucidate a phenotype. As shown, a process 2200 begins by identifying a target protein. See

2204. Then, at 2206, the process involves identifying positive and negative biochemical hits. More generally, this may involve ranking a number of compounds based upon their interaction with the target. In a specific case, the compounds are ranked based upon their binding affinities to or ability to inhibit the enzymatic activity of the target protein.

After the compounds have been characterized in some manner based upon their interaction with the target, they are used to define a cellular phenotype. See 2208. Generally, the techniques to accomplish are the same as described with reference to operation 2006 of Figure 20. In this case however, one may obtain a strong correlation between mechanism of action (involving the target) and phenotype by using multiple ofthe compounds identified at 2206. For example, some ofthe "best hits" may be administered to cell lines in various concentrations. And some of the least effective compounds may also be administered. Cellular attributes that are more strongly exhibited with increasing concentration ofthe best hits (and not exhibited or exhibited only weakly upon administration of the negative hits) can be used to define the virtual phenotype. In a related approach, compounds having widely varying levels interaction with the target are administered to cells. Those cellular attributes that vary linearly or at least monotonically with the degree of interaction between the target and compound represent attributes that can be used to define the virtual phenotype.

After the cellular phenotype has been defined, previously characterized compounds may be clustered with that phenotype. See 2210. As with operation 2110 of Figure 2, this may create or strengthen an association between a mechanism of action and various compounds in a database. Finally, and optionally, procedure 2200 may provide a "higher resolution" mechanism of action for the compounds identified at 2206. See 2212. Presumably interaction with the target suggests a specific mechanism of action or at least some aspect of a mechanism of action. However, a given target may participate in a larger cellular mechanism of action - unknown to researchers. Further, a compound may that binds with the target may participate in multiple mechanisms of action - some of which do not involve the target. By linking the target (and its positive hits) to a particular phenotype, some of these additional cellular level activities can be elucidated. The defined phenotype may have been previously identified as associated with other mechanisms of action or higher resolution mechanisms of action. Thus, the phenotype identified at 2208 can be leveraged to generate a higher resolution mechanism of action at 2212.

As suggested in the above discussion, compounds and associated phenotypes may be stored as database records. Such databases can take on many flavors. In one example, a database includes various pieces of information relevant to oncology. Such database may include numerous compounds classified by cellular phenotype, mechanism of action, toxicity, etc. More specifically, the database may include data on commercially available compounds clustered by cellular phenotypes corresponding to mechanisms of action. Further the databases of interest may extended or combined (via standard relational tables and algebra for example) to include additional data such as pharmacology data, cellular genomics data, gene expression data, protein expression data, etc. In a specific example, the database includes measurements made on a subset ofthe NCI60 cell lines, using DNA, Golgi apparatus, and/or microtubules as markers for defining the phenotypes. Other data includes dosage response information, variation in effect over time, etc. The compounds populating the database could include known National Cancer Institute oncology study compounds. In a specific embodiment, the compound set includes some or all ofthe compounds mentioned in the article "A gene expression database for the molecular pharmacology of cancer," Nature Genetics, 24, pp. 236-244 (March 2000).

Various biological analyses may be conducted to develop additional information for characterizing compound mechanisms of action, etc. For example, a cell count analysis may be used to develop dose response curves, Gl 50 data, etc. The cell cycle may also be analyzed to find out how various stages in the cycle vary in response to particular stimuli. The Golgi apparatus may be analyzed to determine whether it is in a normal state, a dispersed state, a diffused state, etc. As another example, tubulin may be analyzed to determine whether it is normal, de-polymerized, over-polymerized, bundled, etc. Obviously, combinations of such analyses may be performed. For example, properties of the Golgi apparatus or tubulin may be analyzed over one or more cell cycles.

In some embodiments, techniques according to the present invention can provide tools for the later stages of dmg development such as clinical trial design and patient management. The properties of known dmgs, such as clinical trial and patient response information, will be used in a similar fashion as the pre-clinical information to provide predictions about the properties of novel compounds. Because the human cell is the locus of dmg action, a database containing drug-cell interactions will be able to provide predictive value for this aspect of dmg development.

Although the above has generally been described in terms of specific hardware, software, and methods, it is understood that many alternatives can exist. In particular, the present invention is not limited to a particular kind of data about a cell, but can be applied to virtually any cellular data where an understanding about the workings of the cell is desired. Thus, in some embodiments, the techniques ofthe present invention could provide information about many different types or groups of cells, substances, and genetic processes of all kinds. Of course, one of ordinary skill in the art would recognize other variations, modifications, and alternatives. Some examples according to the present invention are provided below.

EXPERIMENTS

To prove the principle and demonstrate the objects of the present invention, experiments have been performed to determine the effects of manipulations on cell stmcture using imaging and analysis techniques applied to a variety of situations. These experiments were performed by growing multiple cell lines in the presence of multiple compounds, or substances. Cells were fixed and stained with fluorescent antibodies or labels to multiple cellular portions . One or more images of the cells were then obtained using a digital camera. Descriptors were built by quantifying and/or qualifying patterns of one or more feature from each image in the cell lines under study. A database was built from the descriptors. As the database grows, it should be able to predict the mechanism of action of an unknown dmg by comparing its effect with the effects of known compounds or to identify data clusters within large libraries of compounds. In a first experiment, an automated method to count the number of cells and differentiate normal, mitotic, and apoptotic cells was created. Approximately, 5,000 HeLa cells were plated per well in a 96 well plate and grown for 3.5 days. The cells were fixed with -20° MEOH for 5 minutes, washed with TBS for 15 minutes, and then incubated in 5 mg/ml Hoechst 33342 in TBS for 15 minutes. Then, 72 images were collected with a 40x objective and 75 ms exposure time.

The analysis was performed on objects that met a certain size criteria that was based on 1) measuring the size of objects in the image that were clearly not cells and 2) excluding the first peak ofthe area histogram (Fig. 8B values 1-4654). Histograms ofthe individual object data were generated for each type of feature. Fig. 8 A shows the histogram for average intensity, and Fig. 8B shows histogram data for the area of each object. Fig. 8C shows the scatter plot ofthe average intensity vs. the area of all ofthe objects. The pattern ofthe scatter plot showed an interesting pattern: a large cluster of cells in one region of the graph, with a scattering of object points in other regions. Because mitotic stmctures are identified as particularly bright objects, most likely due to the biological fact that the chromatin is condensed, the original Hoechst images could be used to identify which cells were either undergoing mitosis, or otherwise looked abnormal. Manual inspection of 917 cells resulted in the classification of each object. Fig. 8D shows a graph where each type of cellular classification is delimited. This graph clearly shows that the mitotic nuclei are brighter than the interphase nuclei. Further, the different phases ofthe cell cycle can be separated using these two features. Figs. 8E-8F show bar graphs ofthe average and standard deviations ofthe areas and average intensities for each cell classification type. These graphs show that interphase nuclei are statistically less bright than mitotic nuclei and that telophase nuclei are statistically smaller than other mitotic nuclei.

Each image was thresholded to an intensity level of 20. A standard area value was set at 9500 pixels. Automated information gathering about all ofthe objects was done and collected into an Excel spreadsheet (for more information see, section on imaging system). The following information was recorded:

IMAGE NAME

OBJECT # AREA

STANDARD AREA COUNT

PERIMETER

FIBER LENGTH

FIBER BREADTH

SHAPE FACTOR

ELL. FORM FACTOR

INNER RADIUS

OUTER RADIUS

MEAN RADIUS

AVERAGE INTENSITY

TOTAL INTENSITY

OPTICAL DENSITY

RADIAL DISPERSION

TEXTURE DIFFERENCE MOMENT

EFA HARMONIC 2, SEMI-MAJOR AXIS

EFA HARMONIC 2, SEMI-MINOR AXIS

EFA HARMONIC 2, SEMI-MAJOR AXIS

ANGLE

EFA HARMONIC 2, ELLIPSE AREA

EFA HARMONIC 2, AXLAL RATIO

EFA HARMONIC 3, SEMI-MINOR AXIS

The following results were obtained:

1,250 objects were counted

201 of those objects has standard area counts > 2 (area > 19000 pixels)

195 objects had areas < 6000 pixels

1529 objects estimated in total

1328 object areas are > 6000 pixels

The data was reduced to 917 objects that were 6000 < area < 19000

For the 917 objects a scatter plot of area vs. average intensity and a histogram ofthe average intensity were generated. • 116 objects that had average intensity intensities > 60 were manually looked at to determine their morphology.

• Of those 116 objects:

6 were dead or indistinguishable 4 were interphase

30 were prophase 32 were metaphase 24 were anaphase 20 were telophase (10 pairs)

• 12 prophase objects were missed because of gray scale cut off. (8 of those prophase cells had gray scale values > 57, as did 7 interphase)

• 1 telophase object was missed because it was too small (<6000)

• 1 prophase object was missed because it was too big (>1900) • 16 mitotic objects were missed because they were parts of objects with standard count > 2.

In sum, out of 917 single objects, the analysis correctly identified 106 out of 130 mitotic objects, or (81% predictive, 91% of identified mitotics). Out of 917 single objects, the analysis incorrectly identified only 10 non-mitotics as mitotics (1% total, 8% of identified mitotics); 14 mitotics as interphase (1.4% total, 1% interphase). An automated classification system that would automatically assign values to each object using these or other measurement features can thus be developed, utilizing the principles set forth herein.

In a second experiment, the effects of Taxol on MDCK cells and the different types of morphological effects were observed. A plurality of MDCK cells grown in 96 well plates were treated with Taxol for 4.5 hours at different concentrations (10 uM-lpM). They were then fixed, labeled with Hoechst, and imaged.

This experiment used a labeling protocol comprising: MEOH fix at - 20°, Wash in PBS, Block in PBS/BSA/Semm/Triton-X 100, Incubate with 5 μg/ml Hoechst 10 minutes, and wash.

Cells were inspected for different morphologies and manually counted at each different dmg concentration in one well. Fig. 9 shows example images from each dmg concentration and the different types of morphologies and cells are highlighted. Fig. 10 shows the distribution of each morphology within the cell population as a function of drug concentration. The higher the concentration of Taxol, the larger proportion of cells underwent apoptosis, and the fewer number of normal mitotic cells were detected.

In a third experiment, the purpose was to determine whether the automated analysis methods developed in the first experiment can detect differences in Hoechst morphology in the presence of 6 known compounds at one concentration and exposure time in one cell line. In this experiment, HeLa cells were separately treated with 6 compounds with known mechanism of action. The quantitative methods described in the first experiment were applied to the Hoechst images. Approximately 5,000 HeLa cells per well were plated in a Costar black- walled 96 well tissue culture treated plate and left to recover in the incubator for 24 hours. After this time, 10 ug/mL of cytochalasin D (CD), Taxol, hydroxyurea, vinblastine, nocodazole, and staurosporine was added to different wells at a 1 :100 addition in DMSO. The cells were incubated in the presence of dmg for 24 more hours. After 24 hours, the cells were removed and fixed as in the first experiment. Then, 9 images per well were collected ofthe Hoechst staining using a lOx objective.

The low magnification images taken of Hoechst were n through the automated image analysis method described in the first experiment. Plots ofthe average intensity and area were made of each compound. Fig. 11 shows the scatter plots of the compounds. The scatter plots of each compound are visually distinct. For example, cells treated with CD are smaller than control, and cells treated with Hydroxyurea are larger and brighter. Furthermore, the number of cells per well was very different (data not shown).

The effects of different compounds can be clearly and automatically distinguished by identifying changes in cellular morphology. This method can also be used to count adherent cells.

The next experiment was to develop clustering algorithms that assign statistically meaningful values to the representative two dimensional data shown in Fig. 10, and even more complicated clustering of all of the multidimensional data that can be extracted across one, and multiple images.

A fourth experiment was performed to obtain high magnification images of two markers in the presence of dmgs. In this experiment, HeLa cells were treated with 80 generic compounds with known mechanism of action. The quantitative methods described in the first experiment were applied to the Hoechst images.

Approximately 5,000 HeLa cells per well were plated in a Costar black walled 96 well tissue culture-treated plate and left to recover in the incubator for 24 hours. After this time, 10 ug/mL of each compound from the Killer Plate from

Microsource Discovery Systems (Gaylordsville, CT) was added to different wells at a 1 : 100 addition in DMSO. The cells were incubated in the presence of dmg for 24 more hours. After 24 hours, the cells were removed and fixed as in the first experiment. In addition to being labeled with Hoechst 33342 (against chromatin), cells were also labeled with 1 unit of rhodamine-conjugated phalloidin (against actin) for 30 minutes. The 96 well plate was imaged twice. Once, 9 images per well were collected ofthe Hoechst staining using a lOx objective. After this, one image per well of both the phalloidin and Hoechst staining was collected using a 40x objective.

The resulting high magnification images were analyzed qualitatively and distinct pattern differences were detected in both the Hoechst and phalloidin images. Fig. 12 shows three example images from the experiment. The top row is the Hoechst staining, and the bottom row is the phalloidin staining from the same well. The columns show the images from wells treated with just DMSO (control), cytochalasin D, and Colchicine. The moφhology of each marker is different in the presence of each dmg. Interestingly, there is an effect in the moφhology ofthe chromatin in the Hoechst image of cytochalasin D, which directly targets the actin cytoskeleton (and thus there is an expected effect in the phalloidin image). Also, there is an effect on the actin cytoskeleton, compared to control, in the presence of colchicine that directly targets the microtubule network. The low magnification images were analyzed as described in the first experiment, and different patterns were seen in both the average intensity vs. area plots, and in the number of cells per well (data not shown). Thus, changes in patterns of a marker that is "down-stream" from the direct target of a compound are detectable. Automated image analysis protocols for actin and other markers can be developed similarly, again utilizing the principles set forth herein.

A fifth experiment was performed to test quadmple labeling of 9 different cell lines grown in normal conditions. In this experiment, NCI-H460, A549, MDA-MD-231, MCF-7, SK-OV-3, OVCAR-3, A498, U-2 OS, and HeLa cells were plated. Then, the cells were fixed and stained for portions ofthe each cell known as DNA, tubulin, actin, and Golgi.

The following table summarizes the procedures for this experiment:

Cells were plated out at different densities for 48 hours. Cells were fixed and labeled by the above method. Cells were imaged using an automated imaging system that collected 9 images from each marker using a lOx objective. Higher magnification images were collected of a few cells for demonstration puφoses.

In this experiment, each cell line demonstrated different moφhological patterns as determined by phase. For example, A549 cells are much more compacted than OVCAR-3 cells as determined by phase contract imaging (data not shown). The different fluorescent markers showed even bigger differences between different cell lines. Figs. 13 and 14 show 4 panels of each marker for A549 (Fig. 13) and OVCAR- 3 cells (Fig. 14). The markers are Hoechst (upper left), Phalloidin (upper right), Lens culinaris (lower left), and DM la antibody (lower right). The following table summarizes the qualitative differences between these images:

Higher magnification images were taken ofthe OVCAR3 cells. Fig. 15 shows the same markers at 20x, and Fig. 16 shows the markers at 40x. While the highest magnification images show the most detail, these images illustrate that very little moφhological or feature information is lost in the lOx images.

These data exemplify the differences in moφhology seen between different cell types. Thus the automated image analysis software can be customized for each marker in each cell type. Different dmgs should effect these moφhologies differentially.

An automated quantification method for each marker and cell line can be similarly developed.

A sixth experiment was conducted with a more sophisticated software package and to develop more flexible image recognition algorithms. In this experiment, prototype image features extraction was performed using MatLab programming language with image toolbox and SDC moφhology toolboxes. Algorithms are being developed that will automatically identify objects on images and to measure various moφhological and feature parameters of these objects. Many different features for each ofthe cellular markers were acquired.

An example of a MatLab program called "AnalyseDNA" that takes as an input an unlimited number of images, identifies individual objects in these images based on either their intensities, or based on edge-detection algorithms, and extracts a number of moφhological and intensity characteristics of these objects. A copy of this program follows:

Listing of the AnalyseDNA. program and of some of the supporting subroutines

function files_analysed = AnalyseDNA (filemask, outpath, nx, ny, filter_range, dext , modifier, sfname)

% AnalyseDNA performs measurements on files of DNA images

% VI. EV 2-11-99; 2-15-99; 2-16-99

%

% files_analysed = AnalyseDNA (filemask, outpath, nx, ny, filter_range, dext, modifier, sfname)

% PARAMETERS: % ALL PARAMETERS ARE OPTIONAL

%

% FILEMASK - mask for file names to be analyzed

INCLUDING PATH (for example c : \images\* . tif )

% DEFAULT '.\*.tif (all *.tif files in the current directory) .

%

% OUTPATH - path to a directory where all the output files will be placed.

% DEFAULT - output is saved in the same directory which contains images

%

% NX, NY - number of individual images in montage images along X and Y axes (DEFAULT 1) % FILTER_RANGE - 3 col-wide array (or[]). Specifies how data is filtered when summary is calculated

% this parameter internally is passed to GetDNAData and then to GetSummaryData - see these % functions for details. For example: [2 2 Inf; 6 100

8000] will case all raws of data for which

% values in column 2 are less than 2 and all raws where values in column 6 are less than 100 or

% more than 8000 to be excluded from all calculations of a summary.

% DEFAULT - [] (means do not filter, summarize all data)

%

% DEXT - string. Extension for data files being saved. % DEFAULT ' dat ' ;

%

% MODIFIER - this modifier is 'SUMMARY', summary file is created;

% 'SUMMARY ONLY' - only summary is generated, data for individual files are not saved

%

% sfname - string. File name of a summary file

% DEFAULT ' summary [date] .dat '

% % OUTPUT:

%

% AnalyseDNA works on image files or montages. For each image file it creates a tab-delimits file of measured % parameters of all the objects in the montage with the same base name as a montage file and extension specified % by dext parameter (or .dat by default) and file

' errors [date] .err' - with the list of files that matched the

% filemask but could not be processed. % If 'summary' or 'summary only' modifier is specified, it also creates a single file

' summary [date] .dat' (or

% different extension, if specified by DEXT) which contains summary information for all analyzed files. %

% ALL OUTPUT FILES are saved in a directory specified by OUTPATH parameter

%

% RETURNS *files_analysed* - number of files that have been successfully processed.

%

% Column designations in the output files are described in GetDNAData

% % FILE NAME CONVENTIONS

% AnalyseDNA attempts to identify a number for each file to identify the file in summary output.

% It does that by looking for the first space or underscore, followed by a number and then takes % as many successive numbers as it can find. If it fails to identify a number it assigns a

% default which is -1

% SEE ALSO GetDNAData, GetSummaryData %

% TO DO improve error handling in opening and writing files (GLOBAL error file ?) % include procedures for writing text headers into the output files

if nargin > 8 error ('Wrong number of input parameters') ; end if nargout >1 error ('Wrong number of output parameters: only one allowed ' ) ; end

% set defaults need_summary = 0 ; summary_only = 0 ; use_default_outpath = 0; datestring = datestr (floor (now) ) ; if nargin == 7 % set default summary file name sfname = ['summary' deblank (datestring) ] ; % extension will be appended later based on dext if deblank (upper (modifier) ) == 'SUMMARY' need_summary = 1 ; elseif deblank (upper (modifier) ) == 'SUMMARY ONLY' need_summary = 1 ; summary_only = 1 ; else error (['Wrong parameter: unknown modifier ' modifier] ) ; end end

if nargin == 5

% default data file extension set dext = ' dat ' ; end i f nargin = = 4

% default filter range filter_range = [] ^■ end if nargin == 3 ny = 1; % default number of images in montage along Y end if nargin == 2 nx = 1 ; end if nargin == 1 use_default__outpath = 1; end if nargin == 0 filemask = '*.tif ' end

% check parameters if ( -ischar (filemask) | -ischar (dext) | -ischar (sfname) ) error ( ' Wrong parameter type: filename, filepath, dext and sfname should be strings') ; end if ( ( size(nx) -= [1 1] ) | ( size (ny) ~= [1 1] ) ) error ('Wrong parameter type: nx and ny should be scalars (lxl arrays) ') ; end if (-isempty (filter_range) & size (filter_range, 2) ~= 3) error ('Wrong parameter type: filter range should be [] or 3 - cols-wide array') ; end

% end testing parameters

% Generate list of files to process datapath = getpath (filemask) ; if use_default_outpath == 1 outpath = datapath; end if exist (outpath, 'dir') ~= 7 error ([' Path ' outpath, 'not found. Exiting..']); elseif exist (datapath, 'dir') ~= 7 error ([' Path ' datapath, 'not found. Exiting..']); end

sfname = makefullname (outpath, sfname, dext) ; if need_summary == 1 if exist (sfname , 'file') disp(['File ', sfname, 'already exists!']); input ('Press ^ΛC to abort, Enter to delete and continue ' ) ; delete (sfname) ; end end

fiist = FileList (getfname (filemask) , datapath); numfiles = size(flist, 1); % total number of files to process disp([ 'About to process ', num2str (numfiles) , ' files'] %DEBUG - commented out "input" to run from Wrod input (' Press ^ΛC to abort, Enter to continue');

% main loop where the job gets done: error_file = makefullname (outpath, ['error' datestring ' • err ' ] ) ; num_processed = 0; num_error =0; for i = 1: numfiles

% first generate file name for a data output file current_fullname = fiist (i, :) ; % full name with path and extension current_datafile = makefullname (outpath, makefname (getbasefname (current_fullname) , dext) ) ;

%extract number from a filename fnumber = getfilenumber (current_fullname) ;

% load an imagefile, record errors read_error = 0 ; try

I = imread (current_fullname) ; %DEBUG disp (['Image file #', num2str (fnumber) , ' loaded ' ] ) ; catch

% record file-opening error in an error_file read_error = 1 ; num_error = num_error +1; msg = [current_fullname ' : ' lasterr] ; add_error_msg (error_file, msg); end

% extract and write data to a file in outpath if read_error ~=1 if (need_summary == 0) %DEBUG disp (['Starting analysis of file #', num2str (fnumber) , '.']); current_data = GetDNAData (I , nx, ny, fnumber) ; %DEBUG disp (['Finished analysis of file #', num2str (fnumber) , '.']);

%load current_data .mat ' current_data ' ; write_data (current_data, current_datafile) ; else %summary needed %DEBUG

[current_data, current_summary] = GetDNAData (I , nx, ny, fnumber, filter_range) ;

%load current_data .mat ' current_data ' ; %load current_summary . mat ' current_summary ' ; write_summary (current_summary, sfname) ; if summary_only ~= 1 write_data (current_data, current_datafile) ; end end end end % of the main for loop num_processed = numfiles - num_error;

|.-=-..=.-==-==---:==-=-end function AnalyseDNA ()

function result = add_error_msg (filename, msg) % adds string MSG to an errorfile FILENAME % returns 1 if success, 0 if failure

err_FID = fopen (filename, 'at'); if err_FID == -1 warning ([' Can not open error file ' filename]); else fprintf (err_FID, '%s\n', msg); fclose (err_FID) ; end

%======================end function add_error_masg ( )

function N = getfilenumber (fname)

% returns the first number extracted from a file name (string) or -1 if fails to extract any number numbers = NumbersFromString ( getfname (fname) ) ; % vector of all numbers encoded in the name

% (but not the path, even if present) if isempty (numbers)

N = (-1) ; % return -1 if no numbers found in the name else

N = numbers (1) ; end

%===================== end function getfilenumber ( )

function result = wπte_data (data_array, file_name) % writes data in a data_array in a tab-delimited ascii file. % result is 0 if success and -1 if failure % if file_name exists, overwrites it result = -1; try fid = fopen (file_name, ' wt ' ) ; i f f id ~ = - 1 for k = 1 : size (data_array, 1) fpπntf (fid, ' %g\t ' , data_array (k, :) ); fprintf (fid, ' \n' ) ; end test = fclose(fιd); result = -1; catch result = -1; end

%===================== end function write data()

function result = wrιte_summary (s_vector, fιle_name) % appends summary vector s_vector to a fιle_name (ASCII tab-delimited file) . % if fιle_name does not exist, creates it. % result is 0 if success and -1 if failure

result = - 1 ; try % debug fid = fopen (file name, 'at'); result = fprintf (fid, ' %g\t ' , s_vector) ; result = fprintf (fid, ' \n'); result = fclose (fid) ; result = 0 ; catch result = -1; end % ======--=======----===-- end function wrιte_summary ( )

function Data = GetOb ectsData (I , Ilabel) % GetObj ectsData returns array measurements of objects image "I" masked by "Ilabel"

% EV 2-3-99; 2-10-99

% OData = GetObj ectsData (I , Ilabel) returns an array of morphological and intensity measurements % taken from a grayscale image "I". Objects are identified on a mask image Ilabel, usually

% created by bwlabelO

% OUTPUT:

% Each row m the output array OData represents individual object

% columns contain the following measurements:

%

% 1 - Index ("number" of an object) ; 8

Solidity; % 2 - X coordinate of the center of mass; 9 - Extent;

% 3 - Y coordinate -"- ; 10 - Total

Intensity;

% 4 - Total Area (m pixels) ; 11 - Avg .

Intensity; % 5 - Ratio of Maj orAxis/MmorAxis ; 12 - Median

Intensity;

% 6 - Eccentricity; 13 - Intensity of

20% bright pixel

% 7 - EquivDiameter; 14 - Intensity of 80% bright pixel

O

% For details on morphological parameters see information on MatLab lmfeature ( ) ; % Intensity parameters are either obvious or are documented m comments this file.

% Procedures this file are documented m notebook file "MATLAB Measuring Nuclei (1) l-29-98.doc"

if (nargin ~= 2) error ( ' function requires exactly 2 parameters ' ) ; end if (nargout ~= 1) error ('function has 1 output argument (array X by 14) ') ; end

% finished checking arguments

% first collect morphological parameters m a structure array:

ImStats = lmfeature (Ilabel , 'Area', 'Centroid', 'MajorAxisLength ' , ... 'MmorAxisLength' , 'Eccentricity', ' EquivDiameter ' ,

'Solidity¹, 'Extent', 8 );

% now convert it into array (matrix) while collecting intensity data for each object:

%preallocate output array: numobjects = size (ImStats , 1) ; OData = zeros (numobjects , 14); %now convert ImStats into array and add intensity data to it for k=l umobj ects

OData (k, 1) = k;

OData(k, 2) = ImStats (k) . Centroid ( 1) ; OData(k, 3) = ImStats (k) . Centroid (2 ) ;

OData(k, 4) = ImStats (k) .Area ;

OData (k, 5) = (ImStats (k) .MajorAxisLengtn) / (ImStats (k) . MmorAxisLength) ; OData (k, 6) = ImStats (k) .Eccentricity ;

OData(k, 7) = ImStats (k) . EquivDiameter ;

OData(k, 8) = ImStats (k) . Solidity ;

OData(k, 9) = ImStats (k) .Extent ;

% now collect and assign intensity parameters from image I

object_pιxels = fmd( Ilabel == k) ; object_area = size (ob ect_pιxels , 1) ; %same as total number of pixels m the object object_mtensιtιes = double ( I (obj ect_pιxels) ) ; % need to convert to double to do math sorted_mtensιtιes = sort (obj ect_mtensιtιes) ; % will need to get median, 20% and 80% pixels total_mtensιty = sum (object_mtensιtιes , 1) ; avg_mtensιty = total_mtensιty / obj ect_area; medιan_mtensιty = sorted_mtensιtιes ( f loor ( obj ect_area/2 ) + 1 ) ; pιx20 = sorted_mtensιtιes ( f loor (object_area*0.2 ) +1 ) ; %bπghtest pixel among dimmest 20% pιx80 = sorted_mtensιtιes ( f loor (obj ect_area*0.8) +1

OData (k, 10) = total_mtensιty ; OData (k, 11) = avg_mtensιty;

OData (k, 12) = medιan_mtensιty ;

OData (k, 13) = pιx20; %brιghtest pixel among dimmest 20% OData (k, 14) = pιx80; %dιmmest pixel among brightest 20% end %for

%--==--------====--------=--== end function

GetObj ectsData () ==----------=====--=====

function Imask = MaskDNAl (I); % MaskDNAl - generates binary mask for cell nuclei through edge detection

% EV 1-22-99; 2-6-99; 2-10-99

% Imask = MaskDNAl (I)

% PARAMETERS % I - intensity image (grayscale)

% OUTPUT

% Imask - BW image with objects from I

%

% For more details see Notebook Matlab_DNA_maskmgl_l-22- 99.doc

% Uses SDC Morphology Toolbox V0.7

if (nargin ~= 1) error (' Wrong number of input parameters'); end if (nargout ~= 1) error ('Wrong number of output parameters: one output argument should be provided'); end

Imask = edge (I, 'canny'); Imask = mmdil (Imask, mmsecross (1) ) ; Imask = mmero ( mmclohole (Imask, mmsecross (1) )) ; Imask = mmedgeoff (Imask, mmsecross (1) ) ;

% note that mmedgeoff this command removed FILLED OBJECTS but not touching OUTLINES .

% these outlines can be removed by filtering: Imask = medfilt2 (Imask, [5 5]) ;

%==--=--====end MaskDNAl --===----=--===--=---========--==========

Given the list of image files or montages of images as an input, this program creates an individual file for each image that contains the following quantitative measurements for all objects identified in the image:

1 - Index ("number" of an object); 8 - Solidity;

2 - X coordinate of the center of mass; 9 - Extent; 3 - Y coordinate -"- ; 10 - Total Intensity;

4 - Total Area (in pixels); 11 - Avg. Intensity;

5 - Ratio of MajorAxis/MinorAxis; 12 - Median Intensity;

6 - Eccentricity; 13 - Intensity of 20% bright pixel

7 - EquivDiameter; 14 - Intensity of 80% bright pixel A fragment of an output for a single file, containing 9 images of cells stained for DNA and acquired with a lOx objective. A montage image that was used as a source to generate data in A is presented in Fig. 17.

The same program also summarizes measurements across many files and performs statistical analysis ofthe summary data. It creates a summary file with the following data:

1 - Image file number;

2 - Average object Area (in pixels); 3 - STD (standard deviation) of 2; 4 - Avg. of Ratio of MajorAxis/MinorAxis; 5 - STD of 4;

6 - Avg. Eccentricity; 7 - STD of 6;

8 - Avg. EquivDiameter; 9 - STD of 8;

10 - Avg. of Solidity; 1 1 - STD of 10; 12 - Avg. of Extent; 13 - STD of 1 1

14 - Avg. of objects Total Intensity; 15 - STD of 14

16 - Avg. of objects Avg Intensity; 16 - STD of 15

18 - Avg. of objects Median intensity; 19 - STD of 18 20 - Avg. of objects intensity of 20% bright pixel; 21 - STD of 19

22 - Avg. of objects intensity of 80% bright pixel; 23 - STD of 21

An example of summary output obtained by running AnalyseDNA against 10 montage files also is shown in Appendix B. A seventh experiment was conducted in order to use sequence analysis algorithms to analyze features of cell images. In this experiment, HeLa cells were treated for 24 hours with several different compounds, and then fixed, and stained with a fluorescent DNA dye. One image of these cells was acquired for each of the treatments and moφhometric parameters and features were measured: Resulting measurements were arranged into a string of numbers and reduced to a pseudo- nucleic acid sequence using following mles: At any given position in the sequence a number was substituted by "t" (a code for thymidine) if its value is among highest 25% ofthe values at the coπesponding position in the data set, "g" if it is between 50% and 25%, "c" if it is between 75% and 50%, and "a" if it belongs to lowest 25%> of values. Thus one descriptor or sequence was generated per treatment as illustrated in Fig. 18.

Resulting sequences were clustered using an AlignX module commercial software package Vector NTI (http://infonTiaxinc.com), which uses a Neighbor Joining algorithm for sequence clustering. The resulting dendrogram is presented in Fig 18. On the dendrogram the closest "leafs" correspond to the closest pseudo-sequences. Interestingly, compounds with similar mechanisms of action cluster together on the dendrogram. Another example ofthe generation of pseudo-sequences and clustering is shown in Fig. 19. In some embodiments, techniques according to the present invention can provide tools for the later stages of dmg development such as clinical trial design and patient management. The properties of known dmgs such as clinical trial and patient response information will be used in a similar fashion as the pre-clinical information to provide predictions about the properties of novel compounds. Because the human cell is the locus of dmg action, a database containing dmg-cell interactions can be able to provide predictive information for this aspect of dmg development. Although the above has generally described the present invention according to specific systems, the present invention has a much broader range of applicability. In particular, the present invention is not limited to a particular kind of data about a cell, but can be applied to virtually any cellular data where an understanding about the workings ofthe cell is desired. Thus, in some embodiments, the techniques ofthe present invention could provide information about many different types or groups of cells, substances, and genetic processes of all kinds. Of course, one of ordinary skill in the art would recognize other variations, modifications, and alternatives.

EV Table 1 doc

Example of the output of AnalyseDNA. m program (measurements for a single 3 by 3 montage image)

Hi on

10 1(11 tisa

Page 1 of 16

EV Table 1 Joe

)(

)(

(( O 0 )(

)(

))

nma

)() )(

na 1(2

13

(((

)(

Page2 of16

EV Table 1 doc

(((

3o BOD

mm

li )(

)•

)( !(

Page 3 of 16

EV Table I doc

( α

)

)(

( net

(

(

)))

100

(

)

)

)

)

• cum )

)

Page 4 of16

EV Table 1 doc

It

(9 in i

)> )) last ιoa

i m

PageS of16

EV Table I doc

)))

sl )(•

•cc

)••

in )(

((((()

(

a

Page 6 of16

EV Table I doc

lo

))

() not ((

)•

• ((((()

))

nun

)(

> (urn

Page 7ofl6

EV Table I doc

)oe oi2 ic π

Page 80.16

EV Table 1 doc

a

)))!

O St o

It

())(

Page9 of16

EV Table I doc tio oic

))) ••mi lcosoi no

!)( non

)))

((> cant

()

))

))!!))

)

oc

Page 10 of 16

EV Table 1 doc tut too

)) so

))( )• o nun

)•

o

))()

)))

)(

O )) so ro

) o •Una

la not

))>

9 non sus ))()

)(

)•

Page II of 16

gllll_jϊlIiϊllIillllallli IIIIlϊllIIlP iiijijiiijjjjijiiij

^■a

> u . EVTablel.doc

O 4

EVTablel.doc

Page 14 of 16

EV Table I doc

so

Page IS of16

Page 16 of 16

EVTable 2.doc

Example of the a-uαmary output of AnalyβθDMA. m program (sumαairy for 10 3 by 3 montaga image a)

o

09

Page 1 of 1

Claims

CLAIMS What is claimed is:

1. A method of predicting a property of a manipulation of cells based upon a descriptor, said method comprising: providing a plurality of cells; manipulating said plurality of cells; capturing a morphological value from said plurality of cells; assigning a degree of presence of said moφhological value; and storing said morphological value and said degree of presence; wherein said descriptor is derived from a first component of a cell and a second component of said cell, said capturing said morphometric value from said plurality of cells comprises determining a relationship between said first component and said second component. 2. The method of claim 1 wherein said first component and said second component are selected from a protein, a protein modification, a nucleic acid, a lipid, a carbohydrate, a subcelluar structure and an organelle.

3. The method of 1 wherein said step of manipulation occurs in a manner selected from a electrical source, a chemical source, a thermal source, a gravitational source, a nuclear source, a temporal source, and a biological source

4. The method of claim 3 wherein said chemical source is selected from a pharmacological candidate and a drug screening library.

5. The method of claim 1 wherein said moφhological value is selected from a count, an area, a perimeter, a length, a breadth, a fiber length, a fiber breadth, a shape factor, a elliptical form factor, an inner radius, an outer radius, a mean radius, an equivalent radius, an equivalent sphere volume, an equivalent prolate volume, an equivalent oblate volume, an equivalent sphere surface area, an average gray value, a total gray value, and an optical density.

6. The method of claim 1 wherein said degree of presence is multiple of a quantized value.

7. A computer program product for populating a database with manipulated biological information, said computer program product comprising: code for providing a plurality of cells in various stages ofthe cell cycle, said stages of the cell cycle including at least one selected from inteφhase, GO phase, Gl phase, S phase, G2 phase, M phase, prophase, prometaphase, metaphase, anaphase, and telophase; code for manipulating said cells in said various stages of cell cycle development to form a plurality of manipulated cells; code for capturing an image of said plurality of manipulated cells; code for determining a descriptor from said image for said manipulated cells; code for populating a database with said descriptor; wherein said image includes a first component of a cell and a second component of said cell; and a computer readable storage medium for holding the codes.

8. The computer program product of claim 7 wherein said first component and said second component are selected from a protein, a protein modification, a nucleic acid, a lipid, a carbohydrate, a sub-cellular structure and an organelle. 9. The computer program product of claim 7 wherein said image is a digitized representation of said plurality of manipulated cells.

11. The computer program product of claim 9 wherein said digitized representation provides a density value of said plurality of manipulated cells.

11. The computer program product of claim 7 wherein said descriptors comprise numeric or logical values.

12. The computer program product of claim 11 wherein said values further comprises a nucleotide.

13. The computer program product of claim 11 wherein said values further comprises an amino acid letter. 14. A system for capturing images of cells or cell structures, the system comprising: a cell holder comprising a plurality of sites in a spatial orientation, each of the sites being capable of holding a plurality of cells to be imaged; an image capturing device coupled to the cell holder, the image capture device being adapted to capture at least one image in at least one ofthe plurality of sites; an illumination apparatus comprising a liquid light guide coupled to the plate for highlighting the plurality of cells in a relatively even spatial manner for image capturing puφoses; an image processing device coupled to the image capturing device, the image capturing device being adapted to convert the image into a digital representation; and a database storage device comprising a database management element coupled to the image capturing device, the database storage device being adapted to retrieve the digital representation of the image from the image processing device and storing the digital representation.

15. The system of claim 14 further comprising a stage comprising a device for moving the cell holder in a spatial direction to traverse across the cell holder in the spatial orientation.

16. The system of claim 14 wherein the illumination apparatus comprises sub-elements, at least one of the sub-elements being positioned away from the image capturing device to prevent a possibility of vibration from the one sub- elements to be transmitted to the image capturing device.

17. The system of claim 14 wherein the digital representation comprises a plurality of regions and objects.

18. The system of claim 14 further comprising a computing device connected between the database storage device and the image processing device. 19. The system of claim 14 wherein the image capturing device comprises a magnification of at least IX and greater to capture the image of the site.

20. The system of claim 14 wherein the plurality of sites comprises at least 96 sites.

21. The system of claim 14 wherein the liquid light guide characterized as a flexible member that substantially prevents vibration from the an element ofthe illumination apparatus to be transferred to the image capturing device.

I l l

22. The system of claim 14 wherein the spatial direction can be selected from an x-direction, a y-direction, or a z-direction in a Cartesian coordinate system.

23. The system of claim 14 wherein the each ofthe sites comprises a volume that is sufficient to prevent a solution therein from evaporating in a substantial manner that may influence the image capturing.

24. A method for identifying a mechanism of action for a first compound, the method comprising the steps of: receiving the first compound; measuring at least one feature of a cellular phenotype to define a target phenotype; identifying additional compounds providing a feature similar to the feature identified in the measuring step; and characterizing the first compound in terms of distance from a specific target phenotype having known characteristics.

25. The method of claim 24 comprising the further step of storing the additional compounds and their associated phenotypes in a database.

26. A method for identifying a mechanism of action for a cellular stimulus, the method comprising the steps of: receiving cells of interest; measuring at least one feature ofthe cells to define and quantify a target phenotype; identifying additional compounds providing a feature similar to the feature identified in the measuring step; and characterizing the first compound in terms of distance from a specific target phenotype having known characteristics.

27. A method for identifying information relevant to at least one of a mechanism of action and cellular activity by utilizing assay data to elucidate a phenotype, the method comprising the steps of: identifying a target protein; identifying positive and negative biochemical hits related to the target protein; defining the target phenotype utilizing the positive and negative hits; identifying other compounds providing similar features; and characterizing the first compound in terms of distance from a specific target phenotype having known characteristics.