Database system for predictive cellular bioinformatics6738716Abstract 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 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 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. Claims What is claimed is: Description COPYRIGHT NOTICE
TABLE
Image Acquisition System Elements
DESCRIPTION MAKER MODEL
Microscope Zeiss 100M
(x-y) motorized stage Prior
Xenon lamp Sutter Lambda
Filter wheel Sutter Lambda-10
Microtitre Plate holder Prior 500-H223R
Isolation Table Kinetic Systems 9101-24-85
Objective Spacers Polytec PI P-721.90
Camera Hamamatsu C47-95
Computer IBM IntelliStation
Software Metamorph v.4
Objectives Zeiss Achroplan 10x/0.25
LD-Achroplan 20x/0.4
LD-Achroplan 40x/0.6
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 of the 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 of the elements of the 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 of the 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 41A 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, Mass. but can also be others. The stage is controlled via control line 61 through controller 43A, 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, Mass. 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 of a 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 structure. The isolation table is designed to prevent excessive vibration of the 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, Mass. 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 Instrument Company of Novato, Calif. 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 48A (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 of the 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 Feb. 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 41C, 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 1.times., 10.times., 20.times., 40.times., 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 epifluorescence, a confocal, a total-internal reflection, a phase, a Hoffmnan, 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. FIG. 6 is a simplified diagram 600 of a cleaning and dispensing system according to an embodiment of the 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. 7A illustrates a representative block flow diagram of simplified process steps of a method for determining properties of a manipulation based upon effects of the 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 of the 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 G0, G1, S, G2 or M phase, M phase including the following cell cycle stages: interphase, prophase, prometaphase, metaphase, anaphase, and telophase. 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 drugs, or also compounds of unknown biological activity. Or exposure to biologics that may or may not be used as drugs such as hormones, growth factors, antibodies, or extracellular matrix components. Or exposure to biologics such as infective materials such as viruses that may be naturally occurring viruses or viruses engineered to express exogenous genes at various levels. Bioengineered viruses 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 centriftigal force. Manipulations could also include changes in gravitational force, including sub-gravitation (the preferred embodiment in outerspace). Manipulations could include application of a 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 structure such as Golgi apparatus, mitochondria, spacing between cell structures or cells themselves, as well as many others. 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 of the descriptors of the manipulation. Then, in a step 708, properties of the manipulation are predicted based upon the properties of the located descriptors. Properties can comprise toxicity, specificity against a subset of tumors, mechanisms of chemical activity, mechanisms of biological activity, structure, 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 of the manipulation to the descriptors of the 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 drugs 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 drugs. 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 of the similarity between descriptors for the drugs in the database. Because the drugs 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 drugs 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 of the 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 >names1<-paste("Cell.line.5", tmp.names, sep=".") >by.compound.matrix<-as matrix (arranged.by.compound) Example of the code for principal component analysis (data preparation) using S language: all.data.princomp<-menuPrincomp(data=by.compound.matrix, scores=T, scores="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(1,2), 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 Wis. 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 of the cells in the presence of the manipulation of step 704 of FIG. 7A 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 which is specific for the portion or portions of interest. A cell portion can include, for example, one or more of the 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 of the 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 of the target directly. Next, in a step 716, the digital representation of the 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 4.pi.* 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 10.times., 20.times., or 40.times. 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 of the 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 of the 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 of the 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 of the 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, Oreg., 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 of the image. A simplified example of a smoothing and background subtraction procedure in a MatLab language is presented in computer code below: function Isubtracted=cmBackgrSubtrl(I,k) % cmBackgrSubtrl(I,k)--simple flat backround (=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=1 % if (nargin ==1) k=1; end if (size(k).about.=1) error(`cmBackgrSubtr1:parameter k should be a number. Exiting . . . `); end % modpixnum=floor(size(I(:),1)/2); % sortedval=sort(double(I(:))); % modpixel=sortedval(modpixnum); modpixel=median(double(I(:))); bg=k*modpixel; Isubtracted=mmsubm(uint8(I), uint8(round(ones(size(I))*k*modpixel))); An example of a procedure for thresholding in computer code (MatLab) is presented below: function thresh=GetThreshByPerim1(I, M) % GetThreshByPerim1(I) Finds optimal thresholding value for image I % N=GetThreshByPerim1(I) Finds thresholding value N for image I % N=GetThreshByPerim1(I, M)--tests threshold values up to M % DEFAULT M=maximum pixel value in I % note that GetThreshByArea is significantly faster % finds a threshold value that causes the maximal change in the % total perimeter of the objects (Russ ????) % see Matlab_Auto_threshold .sub.-- 1-23-99.doc for more details % Note: works somewhat better on SMOOTH images (i.e. medfilt2(I, [3 3]) two times if (nargin==0) error (strcat(mfilename, `:at least one parameter required`)); elseif (nargin==1) M=double(max(I(:))); % test thresholds up to maximum pixel value in I elseif (nargin >2) error (strcat(mfilename, `:too many parameters`)); end if (size(M)>1) error (strcat(mfilename, `:argument M should be a number`)); end Minval=double(min(I(:))); step=1; % generate vertical vector perims with total perimeters of objects at different % threshold values for i=Minval:step:M bwI=im2bw(I, i/255); prI=bwperim(bwI); pr=sum(prI(:)); if (exist(`perims`, `var`)==0) % perims is yet undefined perims=pr; else perims=cat(1, perims, pr); end end % vector prdiffs contains differences between successive perimeters prdiffs=diff(perims); mindecrease=min(prdiffs); minvalues=find(prdiffs==mindecrease); index_of_mindecrease=minvalues(1); thresh=index_of_mindecrease+1; % ==============end GetThresh1================== 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 variety 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=cmMaskDNA1(I); % cmMaskDNA1--generates binary mask for cell nuclei through edge detection % Imask=cmMaskDNA1(I) % PARAMETERS % I--intensity image (grayscale) % OUTPUT % Imask--BW image with objects from I % % For more details see Notebook Matlab_DNA_masking1.sub.-- 1-22-99.doc % Uses SDC Morphology Toolbox VO.7 if (nargin.about.=1) error(`Wrong number of input parameters`); end if (nargout.about.=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 cmMaskDNA1================================= However, embodiments can also use other techniques, such as Fast Fourier Transforms (FFT) and the like as known in the art without departing from the scope of the present invention. (4). In a step 746, a plurality of region features can be determined. For example, in a representative embodiment, image features can include such quantities as area, perimeter, 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=cmGetObjectsData(I, Ilabel) % cmGetobjectsData returns array measurements of objects in image "I" masked by "Ilabel" % EV 2-3-99; 2-10-99 % OData=cmGetObjectsData(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 bwlabel( ) % OUTPUT: % Each row in the output array OData represents individual object % columns contain the following measurements: % % 1--Index ("number" of an object); % 2--X coordinate of the center of mass; % 3--Y coordinate --"--; % 4--Total Area (in pixels); % 5--Ratio of MajorAxis/MinorAxis; % 6--Eccentrisity; % 7--EquivDiameter; 8--Solidity; 9--Extent; 10--Total Intensity; 11--Avg. Intensity; 12--Median Intensity; 13--Intensity of 20% bright pixel 14--Intensity of 80% bright pixel % % For details on morphological parameters see information on MatLab imfeature( ); % Intensity parameters are either obvious or are documented in comments in this file. if (nargin.about.=2) error (`function requires exactly 2 parameters`); end if (nargout.about.=1) error (`function has 1 output argument (array X by 14)`); end % finished checking arguments % first collect morphological parameters in a structure array: ImStats=imfeature(Ilabel, `Area`, `Centroid`, `MajorAxisLength`, . . . `MinorAxisLength`, `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=1:numobjects 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).MajorAxisLength)/(ImStats(k).MinorAxisLength); 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 object_pixels=find(Ilabel==k); object_area=size(object_pixels, 1); % same as total number of pixels in the object object_intensities=double(I(object_pixels)); % need to convert to double to do math sorted_intensities=sort(object_intensities); % will need to get median, 20% and 80% pixels total_intensity=sum(object_intensities, 1); avg_intensity=total_intensity/object_area; median_intensity=sorted_intensities(floor(object_area/2)+1); pix20=sorted_intensities(floor(object_area*0.2)+1); % brightest pixel among dimmest 20% pix80=sorted_intensities(floor(object_area*0.8)+1) OData(k, 10)=total_intensity; OData(k, 11)=avg_intensity; OData(k, 12)=median_intensity; OData(k, 13)=pix20; % brightest pixel among dimmest 20% OData(k, 14)=pix80; % dimmest pixel among brightest 20% end % for ===================end function cmGetObjectsData( )=============================== (5). 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 of the 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. 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 4.pi.* 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 MathWorks, a company with headquarters in Natick, Mass. 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 of the 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 rules 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 are 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 of the 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 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 760, a plurality of cells in various stages of the cell cycle, 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 of the plurality of manipulated cells is captured using image acquisition techniques in order to provide a morphometric characteristic of each of the manipulated cells. (4) In a step 766, an image database may be populated with the image of the plurality of manipulated cells. 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 structures that can mediate most cellular processes, such as cell growth and division, for example. Because the cytoskeleton is a dynamic structure, 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 drugs, 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 drugs, 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 structure, 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 drugs. 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 Wis. 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 of the similarity between descriptors for the drugs in the database. Because the drugs 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 drugs 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 of the 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 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 program Phylip. 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 rule-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 of the 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 structure, mechanism of action, clinical side effects, toxicity, specificity, gene expression, affinity, pharmacokinetics, and the like. The descriptor of a compound of unknown structure from a natural products library could be compared to the descriptors of compounds with known structure and the structure could be deduced from such a comparison. Similarly, such information could lead to better approaches to drug 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 drug 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 drug, 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 drugs can be queried In some embodiments, techniques according to the present invention can provide tools for the later stages of drug development such as clinical trial design and patient management. The properties of known drugs, 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 drug action, a database containing drug-cell interactions will be able to provide predictive value for this aspect of drug development. Although the above has generally been described m 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 of the 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 structure 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 drug 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 create. Approximately, 5,000 HeLa cells were plated per well in a 96 well plate and grown for 3.5 days. The cells were information can be predictive of whether to continue into an animal model for each compound, and which animal model to pursue. In some embodiments, techniques according to the present invention can provide tools for the later stages of drug development such as clinical trial design and patient management. The properties of known drugs, 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 drug action, a database containing drug-cell interactions will be able to provide predictive value for this aspect of drug 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 of the 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 structure 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 drug 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.degree. 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 40.times. 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 of the area histogram (FIG. 8B values 1-4654). Histograms of the individual object data were generated for each type of feature. FIG. 8A shows the histogram for average intensity, and FIG. 8B shows histogram data for the area of each object. FIG. 8C shows the scatter plot of the average intensity vs. the area of all of the objects. The pattern of the 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 structures are identified as particularly bright objects, most likely due to the biological fact that the chromatin is condensed, it seemed reasonable to go back to the original Hoechst images and the identify the cells which 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 exclusively brighter than the interphase nuclei. Further, the different phases of the cell cycle can be separated using these two features. FIGS. 8E-8F show bar graphs of the average and standard deviations of the 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 of the objects was done and collected into an Excel spreadsheet (for more information see, section on imaging system). The following information was recorded:
IMAGE NAME AVERAGE INTENSITY
OBJECT # TOTAL INTENSITY
AREA OPTICAL DENSITY
STANDARD AREA RADIAL DISPERSION
COUNT TEXTURE DIFFERENCE MOMENT
PERIMETER EFA HARMONIC 2, SEMI-MAJOR AXIS
FIBER LENGTH EFA HARMONIC 2, SEMI-MINOR AXIS
FIBER BREADTH EFA HARMONIC 2, SEMI-MAJOR AXIS
SHAPE FACTOR ANGLE
ELL. FORM FACTOR EFA HARMONIC 2, ELLIPSE AREA
INNER RADIUS EFA HARMONIC 2, AXIAL RATIO
OUTER RADIUS EFA HARMONIC 3, SEMI-MINOR AXIS
MEAN RADIUS
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 of the 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). Optionally, the next step is to develop an automated classification system which would automatically assign values to each object using these or other measurement features. 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-1 pM). They were then fixed, labeled with Hoechst, and imaged. This experiment used a labeling protocol comprising: MEOH fix at -20.degree., Wash in PBS, Block in PBS/BSA/Serum/Triton-X 100, Incubate with 5 .mu.g/ml Hoechst 10 minutes, and wash. Cells were inspected for different morphologies and manually counted at each different drug concentration in one well. FIG. 9 shows example images from each drug 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. The next step is to test the automated Hoechst analysis of the first experiment with multiple drugs. 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 drug 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 of the Hoechst staining using a 10.times. objective. The low magnification images taken of Hoechst were run through the automated image analysis method described in the first experiment. Plots of the 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). Based upon the results of this experiment, we conclude that our initial attempts at automatically identifying changes in cellular morphology show that we can clearly distinguish effects of different compounds. 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 drugs. 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, Conn.) was added to different wells at a 1:100 addition in DMSO. The cells were incubated in the presence of drug 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 of the Hoechst staining using a 10.times. objective. After this, one image per well of both the phalloidin and Hoechst staining was collected using a 40.times. 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. Notice that the morphology of each marker is different in the presence of each drug. Interestingly, there is an effect in the morphology of the 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). Based upon the results of this experiment, we conclude that the fact that changes in patterns of a marker that is "down-stream" from the direct target of a compound are detectable illustrates the efficacy of this approach. The next step based upon the results of this experiment is to develop automated image analysis protocols for actin and other markers. A fifth experiment was performed to test quadruple 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 of the each cell known as DNA, tubulin, actin, and Golgi. 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 10.times. objective. Higher magnification images were collected of a few cells for demonstration purposes. In this experiment, each cell line demonstrated different morphological patterns as determined by phase. For example, A549 cells are much more compacted than OVCAR-3 cells as determnined 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 DMI a antibody (lower right). The following table summarizes the qualitative differences between these images:
MARKER A549 OVCAR3
Hoechst/DNA small large
Phalloidin/actin fuzzy crisp - many stress fibers
Lens culinaris/Golgi compact Disperse/punctate
DM1alpha/Tubulin perinuclear evenly distributed
Higher magnification images were taken of the OVCAR3 cells. FIG. 15 shows the same markers at 20.times., and FIG. 16 shows the markers at 40.times.. While the highest magnification images show the most detail, these images illustrate that very little morphological or feature information is lost in the 10.times. images. These data exemplify the differences in morphology seen between different cell types. Thus the automated image analysis software will have to be customized for each marker in each cell type. Different drugs should effect these morphologies differentially. The next steps based upon the results of this experiment are to customize and develop an automated quantification for each marker and cell line. 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 morphology toolboxes. Algorithms are being developed that will automatically identify objects on images and to measure various morphological and feature parameters of these objects. Since at present it is not known which of the measurements will be most useful for subsequent clustering, many different features for each of the 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 morphological and intensity characteristics of these objects. A copy of this program is described in U.S. application Ser. No. 09/310,879, which has been noted. 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);
2 - X coordinate of the center of mass;
3 - Y coordinate -"- ;
4 - Total Area (in pixels);
5 - Ratio of MajorAxis/MinorAxis;
6 - Eccentrisity;
7 - EquivDiameter;
8 - Solidity;
9 - Extent;
10 - Total Intensity;
11 - Avg. Intensity;
12 - Median Intensity;
13 - Intensity of 20% bright pixel
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 10.times. objective. A montage image that was used as a source to generate data presented on FIG. 17. The same program also summarizes measurements across many files and performs statistical analysis of the 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. Eccentrisity;
7 - STD of 6;
8 - Avg. EquivDiameter;
9 - STD of 8;
10 - Avg. of Solidity;
11 - STD of 10;
12 - Avg. of Extent;
13 - STD of 11
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 the aforementioned patent application. 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 morphometric parameters and features were measured: Resulting measurements were arranged into a string of numbers and reduced to a pseudo-nucleic acid sequence using following rules: At any given position in the sequence a number was substituted by "t" (a code for thymidine) if its value is among highest 25% of the values at the corresponding 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 Align)X module commercial software package Vector NTI 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 of the 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 drug development such as clinical trial design and patient management. The properties of known drugs 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 drug action, a database containing drug-cell interactions can be able to provide predictive information for this aspect of drug 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 of the cell is desired. Thus, in some embodiments, the techniques of the 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.
|
Same subclass Same class | ||||||||||