Classifications

• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
  • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
  • G16C20/70—Machine learning, data mining or chemometrics
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
  • G16C20/00—Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
  • G16C20/50—Molecular design, e.g. of drugs

Definitions

• Drug discovery is a long, multiple step process involving identification of specific disease targets, development of an assay based on a specific target, validation of the assay, optimization and automation of the assay to produce a screen, high throughput screening of compound libraries using the assay to identify "hits", hit validation, and hit compound optimization.
• the output of this process is a lead compound that goes into pre-clinical and, if validated, eventually into clinical trials.
• the screening phase is distinct from the assay development phases, and involves testing compound efficacy in living biological systems.
• Drug discovery efforts often lead to identification of bioactive agents that have unknown or only partially understood systemic effects. Determining how these agents act is usually a labor-intensive process with an uncertain conclusion.
• Certain aspects of this disclosure relate to a high-throughput cell-based screening assay that may be employed in drug discovery.
• the screening method comprises contacting test cells with a test compound to provide contacted test cells, obtaining values for a plurality of cytological attributes of the contacted test cells, and scoring the cells using the values to provide a likelihood score for at least one of a plurality of classifiers, where the plurality of classifiers are defined using values for cytological attributes of cells that have been contacted with compounds of known mode of action.
• the method may involve comparing values obtained from the individual cells in the population to a classifier, determining whether the individual cells are classified or are not classified by the classifier, and calculating the likelihood score using the number of individual cells that are classified by the classifier and the number of cells that that are not classified by the classifier.
• a microscopy system comprising a device for capturing an image of a population of cells; and a computer, operably linked to the device, comprising
• programming for: i. analyzing the image to provide values for a plurality of cytological attributes of the cells; and ii. scoring the cells using the values to provide a likelihood score for at least one of a plurality of classifiers, where the plurality of classifiers are defined using values for cytological attributes obtained from cells that have been contacted with
• An image standardization method includes: a) subtracting the median background pixel value of a first image of cells that are present in a first well of a multi-well plate and contacted with a test agent, from the pixel values of the image to provide a second image, and b) dividing the pixel values of the second image by the median foreground pixel values of untreated cells in a second well of the multi- well plate, thereby providing a third image.
• the pixel values of the third image may be rescaled.
• a computer readable medium comprising executable instructions for performing this method is also provided.
• a method for providing a phenotypic classifier comprises: a) contacting a first population of cells with a first compound having a first known mode of action to provide a first population of contacted cells; and b) contacting a second population of cells with a second compound having a second known mode of action to provide a second population of contacted cells; c) obtaining values for a plurality of cytological attributes of the first and second populations of contacted cells as well as an untreated population of cells; and d) identifying ranges of values for each of the cytological attributes that, together, distinguish the first population of contacted cells from the second population of contacted cells and the untreated population of cells.
• FIG. 1 schematically illustrates one embodiment of a method.
• FIG. 2 shows that staining intensity is a major source of assay variation.
• FIG. 3 provides a method by which an image is standardized.
• FIG. 4 illustrates that an offset of 0.01 and a multiple of 3 provides acceptable results.
• FIG. 5 shows that the image standardization method decreases false positive predictions.
• FIG. 6 shows that classifiers can be made more robust by increasing the number of control training sets.
• FIG. 7 schematically illustrates a typical dose calculation.
• FIG. 8 schematically illustrates one embodiment of a method.
• FIG. 9 illustrates a well classification using a Bayesian approach.
• FIG. 10 shows that classifier performance can be assessed by recall and precision metrics.
• FIG. 11 illustrates an exemplary assay.
• FIG. 12 is a graph of exemplary results.
• FIG. 13 is a table of exemplary results.
• FIG. 14 shows graphs illustrating the average recall performance of some classifiers.
• FIG. 15 shows graphs illustrating the average performance of some classifiers.
• FIG. 16 is a table showing exemplary compounds and their mechanism of action.
• FIG. 17 shows heatmaps that reveal phenotypic patterns.
• FIG. 18 shows a close-up of a portion of the heatmap shown in FIG. 17.
• determining means determining if an element is present or not. These terms include both quantitative and/or qualitative determinations. Assessing may be relative or absolute. “Determining the presence of includes determining the amount of something present, as well as determining whether it is present or absent.
• contacting means to bring or put together.
• a first item is contacted with a second item when the two items are brought or put together, e.g., by touching them to each other or combining them in the same solution.
• a cell that is contacted with an agent is a cell in vitro, i.e., a cultured cell.
• Introducing into a cell e.g., introducing a nucleic acid into a cell, is encompassed by the term "contacting".
• candidate agent and “test compounds” are used to refer to an oligonucleotide, polynucleotide, inhibitory RNA (which may be administered as a shRNA), gene product, polypeptide, small molecule, e.g., up to 2500 Daltons (Da) in size, and any pharmacological compound that is combined with cells in an assay to determine if the agent has a biological activity.
• a candidate agent may be delivered as a nucleic acid that is transcribed and/or translated to provide the candidate agent, for example, an inhibitory RNA molecule or a polypeptide.
• cytological attribute refers to a phenotypic attribute of a cell or a subcellular structure thereof, e.g., a cell's nucleus or an actin staining pattern. Size, staining intensity, shape, elipticity and texture are examples of cytological attributes.
• the term "cytological attribute” may be referred to as a "phenotypic attribute", “parameter”, or “feature” in certain other publications (e.g., Young et al, Nature Chemical Biology 2007 4: 59-68; Feng et al, Nature Reviews 2009 8: 567-578). Cytological attributes may be identified by staining. Many examples of cytological attributes are described in reference cited below.
• cytological attribute refers to a numerical evaluation of (e.g., a measurement) of a cytological attribute or a statistical derivative (e.g., the average, median or variation in) of a plurality of numerical evaluations.
• exemplary values for cytological attributes include size measurements for either the cell or nucleus of a cell, which can include measurements of area, length, width, diameter, etc., total, median or the variation in intensity of staining of the cell or nucleus thereof, irregularity in shape, degree of elipticity and texture, etc.
• the number of values obtained for a single cell may be in the range of 20 to 500 or more, depending on the desired level of complexity.
• classifier refers to a collection of ranges of values of cytological attributes that, together, define a phenotype produced by contacting a cell with a bioactive agent. If the bioactive agent has a defined mode of action, the phenotype of the contacted cell, and therefore the classifier, defines the mode of action of the bioactive agent. For example, a particular phenotype that defines a mode of action of a bioactive agent may be defined using ranges of over 100 different values, which ranges distinguish the phenotype of a contacted cell from the phenotype of control cells or other cells that are contacted with other bioactive agents that have a different mode of action.
• likelihood score refers to an estimate of the certainty of a prediction.
• a likelihood score is not binomial. Rather it is a continuously variable number, which may be a ratio, an odds or a scaled number, e.g., a percentage.
• Bayesian theorem is a theorem in which one conditional probability (such as the probability of a hypothesis given observed evidence) depends on its inverse (in this case, the probability of that evidence given the hypothesis).
• Bayesian theory is described in Howson (Scientific Reasoning: The Bayesian Approach 1993 Open Court) and Jaynes (Probability theory: the logic of science 2003. Cambridge University Press), which are both incorporated by reference herein.
• profile of likelihood scores refers to a set of likelihood scores for different classifiers, where each likelihood score provides an estimate of a certainty of the prediction.
• mode of action refers to a specific biochemical interaction through which a bioactive agent produces a pharmacological effect.
• pixel value refers to the intensity of pixel.
• a pixel value may be a natural number in the range of 0 to 65,536.
• a pixel value may be re-scaled to fall in the range of 0-1, e.g., by dividing the pixel value by 65,536 in the case of a 16-bit image.
• plurality refers to two or more, e.g., at least 2, at least 5, at least 10, at least 50, at least 100, at least 1,000, up to 10,000 or 100,000 or more.
• background refers to those parts of the image that correspond to areas between cells.
• the term "foreground” refers to those parts of the image that are within the outer perimeter of each cell in the image.
• reference to “a cell” includes a plurality of cells
• reference to “a candidate agent” includes reference to one or more candidate agents and equivalents thereof known to those skilled in the art
• reference to “a value” includes reference to values that are averaged across two or more samples, and so forth.
• the method generally includes contacting cells grown in culture with a test agent, and then imaging the cells to provide an image.
• the cytological attributes of the cells in the image are analyzed to provide a plurality of values for each of the cells, and the values for each cell are compared to a classifier that either does or does not classify the cell.
• This step for each population of cells, produces a binomial output that indicates whether the individual cells in the population are classified or not classified by the classifier.
• the classifier is defined using values for the same cytological attributes as analyzed for the test cells, except that the values for the cytological attributes are obtained from cells that have been contacted with compound having a known mode of action.
• the classifier distinguishes the phenotype of cells exposed to a biological agent with a known mode of action from other cells. Using the binomial output, as well as an estimate of the performance of the classifier using positive controls, a likelihood score that the test agent has the same mode of action as the agent having a known mode of action is calculated.
• the method may be employed in a "high-throughput" manner in which multiple populations of cells are grown in the wells of a multi-well plate, there is a control on every multi-well plate (e.g., a population of cells that are not contacted with any agent), and the imaging and analysis is done using an automated microscope system.
• This disclosure provides an image standardization method.
• an image of cells in a well of a multi-well plate is used to standardize an image of cells in a different well of the same multi-well plate.
• a first population of cells is cultured in a test well of a multi-well culture plate and contacted with a test agent.
• An image of those cells is then standardized against an image of a population cells grown in a control well that is present on the same plate as the test well.
• the values of the pixels that make up the image of the first population of cells are adjusted.
• the image of the test cells i.e., the "first" image
• the median intensity of the background pixels is calculated.
• the median intensity of the background pixels of the first image is subtracted from the pixel values of the first image to provide a second "background- subtracted" image.
• the pixel values for the second image are then divided by the median foreground pixel values of an image of untreated cells in a second well of the same multi-well plate, thereby providing a third image.
• the cells in the test well are contacted with a test agent that is present in an inert excipient, e.g., water, ethanol or a dipolar aprotic solvent such as DMSO, and the cells in the control well are contacted with only the inert excipient.
• an inert excipient e.g., water, ethanol or a dipolar aprotic solvent such as DMSO
• the cells in the test well are contacted with an agent dissolved in DMSO, and the cells in the control well are contacted with DMSO alone.
• results from multiple control wells may be employed in this method, e.g., by averaging their results.
• the pixel values that make up the third image may be further adjusted so that they are above zero, and so that they are approximately in the same scale as for other images.
• the pixel values that make up the third image may be rescaled to produce a scaled image in which the pixel values equal an offset + pixel values for the third image value/multiple*(l-offset), where the offset raises the values for all of the pixels in the third image above zero and the multiple is at least 1.
• the offset is below 0.1, and the multiple is at least 1 (e.g., in the range of 1-10).
• Fig. 3 Such a method is illustrated in Fig. 3.
• the method may be repeated for images of cells in other wells of the multi-well plate.
• at least some of the cells in the other wells have been contacted with further test agents.
• the method may be used to standardize every test well of a multi- well plate prior to further processing of the images of the test wells.
• a computer readable medium comprising executable instructions for performing this method is also provided. Such a computer readable medium is described in greater detail below.
• a method for defining a phenotypic classifier which, as noted above, is a collection of ranges of values of cytological attributes that, together, define a phenotype produced by a bioactive compound.
• Classifiers when used in a screening method such as that described below, not only allow cells to be distinguished from one another based on their phenotype, but also identify a cell's phenotype as being similar or identical to that of the cells that were used to define the classifier.
• agents can be screened for those that produce a similar fingerprint.
• a compound that has a mode of action that is similar to that of a compound with a known mode of action can be identified.
• This method involves identifying ranges of values for a plurality of cytological attributes for cells that have been exposed to a first bioactive agent, e.g., an agent that has a defined mode of action, where the ranges of values distinguish those cells from other cells that have been exposed to excipient alone and/or cells exposed to other bioactive agents that produce a different phenotype to the first bioactive agent, e.g., agents that have a different mode of action to the first bioactive agent.
• a first bioactive agent e.g., an agent that has a defined mode of action
• the first step of the method involves contacting a first population of cells with a first compound having a first known mode of action to provide a first population of contacted cells, and contacting a second population of cells with a second compound having a second known mode of action to provide a second population of contacted cells.
• Values for a plurality of cytological attributes for the first population of contacted cells, the second population of contacted cells and for control cells that have been exposed only to excipent are obtained from images of the cells, and ranges of values for each of the cytological attributes that, together, distinguish the population of contacted cells from the second population of contacted cells and the untreated population of cells are identified.
• the populations of cells may be on the same or different multi-well plates, and in certain embodiments, the first population of cells and the untreated population of cells are present in a first multi-well plate.
• the method may further include the step of obtaining values for another untreated population of cells grown on a second multi-well plate, and identifying ranges of values for each of the cytological attributes that, together, distinguish the first population of contacted cells from the second population of contacted cells, the untreated population of cells grown on the first multi-well plate, and the second untreated population of cells.
• Further untreated populations of cells, grown on different multi-well plates or grown at different times may also be employed.
• the method may also involve contacting a third population of cells with a third compound having a third known mode of action to provide a third population of contacted cells, obtaining values for the plurality of cytological attributes of the third population of contacted cells, and identifying ranges of values for each of the cytological attributes that, together, distinguish the first population of contacted cells from the second and third populations of contacted cells and the untreated population of cells.
• Further images of populations cells that have been exposed to other bioactive agents with different modes of action may also be employed.
• these cells may be grown on different multi-well plates or grown at different times (e.g., within at least a week, month or year earlier or later than the time at which the first population of cells was grown).
• the classifier becomes more robust as more untreated populations of cells and more populations of cells that have been exposed to bioactive agents having different modes of action are used to build the classifier.
• the classifier for each bioactive agent may be recalculated periodically using new data.
• the bioactive agents that are used to define a classifier are used at concentrations at which they produce a phenotype.
• the bioactive agents may be employed at a concentration that is at or above their EC50.
• bioactive agents that can be employed in this method and their modes of action are shown in Fig. 16.
• chemotherapeutic agents for the treatment of cancer, and anti-inflammatory agents.
• the agent may target a cell surface receptor (e.g., a GPCR or cell surface tyrosine kinase receptor), or a cytoplasmic protein, for example.
• the bioactive agent may be an antisense RNA, or an inhibitory RNA molecule (which may be administered directly to the cell or indirectly to the cell using a vector encoding the RNA, for example).
• Exemplary agents that can be employed in this method include:
• antiproliferative/antineoplastic drugs such as alkylating agents (for example cis- platin, oxaliplatin, carboplatin, cyclophosphamide, nitrogen mustard, melphalan, chlorambucil, busulphan, temozolamide and nitrosoureas); antimetabolites (for example gemcitabine and antifolates such as fluoropyrimidines like 5-fluorouracil and tegafur, raltitrexed, methotrexate, cytosine arabinoside, and hydroxyurea); antitumour antibiotics (for example anthracyclines like adriamycin, bleomycin, doxorubicin, daunomycin, epirubicin, idarubicin, mitomycin-C, dactinomycin and mithramycin); antimitotic agents (for example vinca alkaloids like vincristine, vinblastine, vindesine and vinorelbine and tax
• cytostatic agents such as antioestrogens (for example tamoxifen, fulvestrant, toremifene, raloxifene, droloxifene and iodoxyfene), antiandrogens (for example
• bicalutamide, flutamide, nilutamide and cyproterone acetate), LHRH antagonists or LHRH agonists for example goserelin, leuprorelin and buserelin
• progestogens for example megestrol acetate
• aromatase inhibitors for example as anastrozole, letrozole, vorazole and exemestane
• inhibitors of 5oc-reductase such as finasteride
• anti-invasion agents for example c-Src kinase family inhibitors like 4-(6-chloro- 2,3-methylenedioxyanilino)-7-[2-(4-methylpiperazin-l-yl)ethox- y]-5-tetrahydropyran-4- yloxyquinazoline (AZD0530; International Patent Application WO 01/94341), N-(2-chloro- 6-methylphenyl)-2- ⁇ 6- [4-(2-hydroxyethyl)piperazin- 1 -yl] -2-met- hylpyrimidin-4- ylamino ⁇ thiazole-5-carboxamide (dasatinib, BMS-354825; J.
• anti-invasion agents for example c-Src kinase family inhibitors like 4-(6-chloro- 2,3-methylenedioxyanilino)-7-[2-(4-methylpiperazin-l-yl)ethox
• inhibitors of growth factor function include growth factor antibodies and growth factor receptor antibodies (for example the anti-erbB2 antibody trastuzumab [HerceptinTM], the anti-EGFR antibody panitumumab, the anti-erbBl antibody cetuximab [Erbitux, C225] and any growth factor or growth factor receptor antibodies disclosed by Stem et al. Critical reviews in oncology/haematology, 2005, Vol.
• inhibitors also include tyrosine kinase inhibitors, for example inhibitors of the epidermal growth factor family (for example EGFR family tyrosine kinase inhibitors such as N-(3-chloro-4-fluorophenyl)-7-methoxy-6-(3-morpholinopropoxy)quinazolin-4- -amine (gefitinib, ZD1839), N-(3-ethynylphenyl)-6,7-bis(2-methoxyethoxy)quinazolin-4-amine (erlotinib, OSI-774), and 6-acrylamido-N-(3-chloro-4-fluorophenyl)-7-(3- morpholinopropoxy)-quinazol- in-4-amine (CI 1033), and erbB2 tyrosine kinase inhibitors such as lapatinib); inhibitors of the hepatocyte growth factor family; inhibitors of
• antiangiogenic agents such as those which inhibit the effects of vascular endothelial growth factor, for example the anti- vascular endothelial cell growth factor antibody bevacizumab (Avastin) and for example a VEGF receptor tyrosine kinase inhibitor such as vandetanib (ZD6474), vatalanib (PTK787), sunitinib (SU11248), axitinib (AG013736), pazopanib (GW 786034) and 4-(4-fluoro-2-methylindol-5-yloxy)-6-methoxy-7-(3- pyrrolidin-l-ylpropoxy)- quinazoline (AZD2171; Example 240 within WO 00/47212), compounds such as those disclosed in International Patent Applications W097/22596, WO 97/30035, WO 97/32856 and WO 98/13354 and compounds that work by other mechanisms (for example linomide,
• vascular damaging agents such as Combretastatin A4 and compounds disclosed in International Patent Applications WO 99/02166, WO 00/40529, WO 00/41669, WO 01/92224, WO 02/04434 and WO 02/08213;
• an endothelin receptor antagonist for example zibotentan (ZD4054) or atrasentan;
• antisense therapies for example those which are directed to the targets listed above, such as ISIS 2503, an anti-ras antisense;
• (ix) gene therapy approaches including for example approaches to replace aberrant genes such as aberrant p53 or aberrant BRCA1 or BRCA2, GDEPT (gene-directed enzyme pro-drug therapy) approaches such as those using cytosine deaminase, thymidine kinase or a bacterial nitroreductase enzyme and approaches to increase patient tolerance to
• GDEPT gene-directed enzyme pro-drug therapy
• the bioactive agent used in the method may be an antitumor alkylating agent, antitumor antimetabolite, antitumor antibiotic, plant-derived antitumor agent, antitumor platinum complex, antitumor campthotecin derivative, antitumor tyrosine kinase inhibitor, monoclonal antibody, interferon, biological response modifier, hormonal anti-tumor agent, anti-tumor viral agent, angiogenesis inhibitor, differentiating agent, PI3K/mTOR/AKT inhibitor, cell cycle inhibitor, apoptosis inhibitor, hsp 90 inhibitor, tubulin inhibitor, DNA repair inhibitor, anti-angiogenic agent, receptor tyrosine kinase inhibitor, topoisomerase inhibitor, taxane, agent targeting Her-2, hormone antagonist, agent targeting a growth factor receptor, or a pharmaceutically acceptable salt thereof.
• the anti-tumor agent is citabine, capecitabine, valopicitabine or gemcitabine.
• the agent is selected from the group consisting of Avastin, Sutent, Nexavar, Recentin, ABT-869, Axitinib, Mnotecan, topotecan, paclitaxel, docetaxel, lapatinib, Herceptin, lapatinib, tamoxifen, a steroidal aromatase inhibitor, a non-steroidal aromatase inhibitor, Fulvestrant, an inhibitor of epidermal growth factor receptor (EGFR), Cetuximab, Panitumimab, an inhibitor of insulin-like growth factor 1 receptor (IGF1R), and CP-751871.
• the performance of a classifier may be evaluated by contacting further populations of cells with the test compound (i.e., the same compound as contacted with the first population of cells), obtaining values for the cytological attributes of individual cells in those populations of cells, and determining if the values correctly classify the individual cells.
• the results from these assays can be summed to provide a performance characteristic for that new classifier that indicates the true positive/true negative rate of the classifier.
• some cells in a population of cells exposed to an agent having a known mode of action retain a "wild-type" appearance and may resemble controls that are not contacted with the agent.
• This performance characteristic accommodates for variation in the phenotype in individual cells in a population.
• the method provides a metric of classifier performance, which, as described below, may be employed to calculate a likelihood score using, for example, Bayesian theory.
• a screening method in which a population of cells is contacted with a test agent, values for cytological attributes are obtained, and the values are compared to a classifier in order to determine if the cells can be classified by the classifier.
• the comparison provides a score of the likelihood that the agent produces the same phenotype as that used to produce the classifier.
• the method may further comprise identifying a test compound having a desired mode of action.
• the method involves: contacting a population of test cells with a test compound to provide contacted test cells; obtaining values for a plurality of cytological attributes of the contacted test cells; and scoring the contacted test cells using the values to provide a likelihood score for at least one of a plurality of classifiers, where the plurality of classifiers are defined using values for the cytological attributes obtained from cells that have been contacted with compounds of known mode of action.
• a plurality of assays may be run in parallel with different agent concentrations to obtain a differential response to the various concentrations. The concentrations may be chosen to encompass an agent's predicted EC50. In particular embodiments, the concentration used in the assay that is immediately above an agent's EC50 (i.e., the agent's "EC50+1”) may be used in the method.
• the values may be obtained by capturing an image of the contacted test cells and analyzing the image to provide the values.
• the image may be captured using an automated microscope and the analysis may be done by a computer operably linked to the automated microscope.
• the scoring may be done by comparing values obtained from the individual cells in the population to a classifier; determining whether the individual cells are classified or are not classified by the classifier, thereby providing a binomial output indicating whether the cell is or is not classified; and calculating the likelihood score using the number of individual cells in the population that are classified by the classifier and the number of individual cells in the population that that are not classified by the classifier.
• the scoring employs a Bayesian theory that uses a metric of the performance of the classifier as an input. As explained above, this metric can be
• the likelihood score is calculated by inputting the binomial output of the comparison (which indicates the number of individual cells in the population that are classified by the classifier and the number of individual cells in the population that that are not classified by the classifier), and the performance score of the classifier.
• a population of cells is contacted with a test agent, and the values for the population of cells are compared to at least one classifier (e.g., one classifier, two or more classifiers, or all classifiers) of a plurality of different classifiers, where each classifier is determined using an agent having a known mode of action.
• the values may be compared to at least two, at least 5, at least 10, at least 20, up to 50 or 100 or more classifiers, where each of the classifiers is determined using a different agent having a known mode of action.
• the method may be employed to produce, for each test compound, a likelihood score for each of a plurality of different classifiers.
• the profile of likelihood scores may be cross-compared with one another to identify agents that have similar likelihood score profiles.
• Such a hierarchical cluster analysis method may be adapted from the methods generally described in Eisen (Cluster analysis and display of genome-wide expression patterns Proc. Natl. Acad. Sci. 1998 95: 14863-14868) and Ling (A computer generated aid for cluster analysis.
• the cluster analysis may be used to generate a so called "heat map", i.e., a two dimensional graphical representation of data where the likelihood scores are represented by different colors and/or different intensities, where the compounds are listed in one dimension and classifiers in the other.
• a tree map may also be generated.
• test agents may be clustered by their mode of action, and test agents having a similar likelihood score profile to an agent with a known mode of action may be identified.
• test compounds may provide a new pattern of scores for a plurality of classifiers, thereby indicating that the test compounds have a third mode of action.
• a test compound may be strongly or intermediately positive for a combination of two or more classifiers, in which case the test compound may have a mode of action that is different to those used to define the classifiers.
• compounds having a mode of action that is different to those used to define the classifiers may still be identified.
• the new mode of action may be identified only after a number of different test compounds have been assayed and a pattern that is consistently different to the patterns produced by the compounds of known mode of action has been identified.
• a test compound with a different pattern may be tested to further define the mode of action of that compound.
• Classifier performance may also be evaluated using other statistical means, e.g., using precision (which is a measure of exactness, i.e., how frequently the method produces false positives and false negatives) and recall (which is a measure of completeness, i.e., how well the method identifies desired compounds) metrics, as illustrated in Fig. 10.
• precision may be defined as the number of items correctly labeled as belonging to the positive class divided by the total number of elements belonging to the positive class
• recall may be defined as the number of true positives divided by the total number of elements that actually belong to the positive class.
• test agent has an unknown mode of action.
• the test agent may be a bioactive agent or a derivative thereof, identified using a different screen, where the term "agent” as used herein describes any molecule, e.g. protein or non-protein organic or inorganic compound.
• Test agents encompass numerous chemical classes, e.g., synthetic, semi-synthetic, or naturally-occurring inorganic or organic molecules.
• Candidate agents include those found in large libraries of synthetic or natural compounds. For example, synthetic compound libraries are commercially available from Maybridge Chemical Co. (Trevillet, Cornwall, UK), ComGenex (South San Francisco, CA), and
• Candidate agents may be small organic or inorganic compounds having a molecular weight of more than 50 and less than about 2,500 Da.
• Candidate agents may comprise functional groups necessary for structural interaction with proteins, particularly hydrogen bonding, and may include at least an amine, carbonyl, hydroxyl or carboxyl group, and may contain at least two of the functional chemical groups.
• the candidate agents may comprise cyclical carbon or heterocyclic structures and/or aromatic or polyaromatic structures substituted with one or more of the above functional groups.
• Candidate agents are also found among biomolecules including peptides, saccharides, fatty acids, steroids, purines, pyrimidines, derivatives, structural analogs or combinations thereof.
• Candidate agents are obtained from a wide variety of sources including libraries of synthetic or natural compounds. For example, numerous means are available for random and directed synthesis of a wide variety of organic compounds and biomolecules, including expression of randomized oligopeptides. Alternatively, libraries of natural compounds in the form of bacterial, fungal, plant and animal extracts are available or readily produced.
• natural or synthetically produced libraries and compounds are readily modified through conventional chemical, physical and biochemical means, and may be used to produce combinatorial libraries.
• Known pharmacological agents may be subjected to directed or random chemical modifications, such as acylation, alkylation, esterification, amidification, etc. to produce structural analogs.
• New potential therapeutic agents may also be created using methods such as rational drug design or computer modeling.
• Screening may be directed to known pharmacologically active compounds and chemical analogs thereof, or to new agents with unknown properties such as those created through rational drug design.
• Agents that modulate a phenotype may decrease the phenotype by at least 10%, at least 20%, at least 30%, at least 40%, at least 50%, at least 60%, at least 70%, at least 80%, or at least 90%, or more, relative to a control that has not been exposed to the agent.
• Agents of interest may be subjected to directed or random and/or directed chemical modifications, such as acylation, alkylation, esterification, amidification, etc. to produce structural analogs.
• Such structural analogs include those that increase bioavailability, and/or reduced cytotoxicity.
• Those skilled in the art can readily envision and generate a wide variety of structural analogs, and test them for desired properties such as increased bioavailability and/or reduced cytotoxicity, etc.
• the cultured cell employed in the assay may be any cell, including immortalized cells and inflammatory system cells which can be screened to identify anti-cancer and antiinflammatory agents, respectively.
• Cultured cells from any animal e.g., cultured mammalian cells, may be employed, including but not limited to: monkey kidney cells (COS cells), monkey kidney CVI cells transformed by SV40 (COS-7, ATCC CRL 165 1); human embryonic kidney cells (HEK-293, Graham et al. J. Gen Virol. 36:59 (1977)); baby hamster kidney cells (BHK, ATCC CCL 10); Chinese hamster ovary-cells (CHO, Urlaub and Chasin, Proc. Natl. Acad. Sci.
• mice Sertoli cells TM4, Mather, Biol. Reprod. 23:243-251 (1980)
• monkey kidney cells CVI ATCC CCL 70); african green monkey kidney cells (VERO-76, ATCC CRL-1587); human cervical carcinoma cells (HELA, ATCC CCL 2); canine kidney cells (MDCK, ATCC CCL 34); buffalo rat liver cells (BRL 3A, ATCC CRL 1442); human lung cells (W138, ATCC CCL 75); human liver cells (hep G2, HB 8065); mouse mammary tumor (MMT 060562, ATCC CCL 51); TRI cells (Mather et al., Annals N. Y. Acad.
• the cultured cell may be a cultured myocyte, e.g., a cultured cell of skeletal muscle, smooth muscle, or cardiac muscle origin. Methods for culturing such cells are known.
• the method may be used to identify an agent that does not produce "side-effects" e.g., undesirable phenotypic changes to a cell.
• a test agent having a desired mode of action has a profile of likelihood scores that is similar to those of an agent of known mode of action. Any agent identified by the above-described method may be tested in a further in vitro assay or using an animal model prior to a clinical evaluation.
• a microscopy system comprises: a device for capturing an image of a population of cells (which may contain a digital camera (e.g., a CMOS camera), an appropriate light source (e.g., a lasers, etc.) and a optical system that may include a beam splitter, a polarizer, a prism, a filter and lenses for transporting light from the light source to the population of cells and for transporting light from the cells to a detector); and a computer, operably linked to the device via, e.g., a cable or wireless connection, that contains programming for: i. analyzing an image of cells to provide values for a plurality of cytological attributes of the cells in the image; and ii.
• a digital camera e.g., a CMOS camera
• an appropriate light source e.g., a lasers, etc.
• a optical system that may include a beam splitter, a polarizer, a prism, a filter and lenses for transporting light
• the device of the microscopy system may be an automated microscope.
• a physical memory of the computer contain a physical computer- readable medium containing instructions (i.e. "programming") for performing the method described above.
• the programming can be provided in a physical storage or transmission medium.
• a computer receiving the instructions can then execute the algorithm and/or process data obtained from the subject method.
• Examples of storage media that are computer-readable include floppy disks, magnetic tape, DVD, CD-ROM, a hard disk drive, a ROM or integrated circuit, a magneto-optical disk, or a computer readable card such as a PCMCIA card and the like, whether or not such devices are internal or external to the computer.
• a file containing information can be "stored” on computer readable medium, where "storing” means recording information such that it is accessible and retrievable at a later date by a computer on a local or remote network.
• data from the microscope is collected, and programming containing the classifier is executed.
• programming containing the classifier is executed. The method described above can be executed
• Active compounds are suspended in DMSO to a concentration of lOmM from powder.
• Tumor cell lines were obtained from ATCC and cultured using the recommended media. Cell splitting was done with calcium and magnesium-free Phosphate Buffered Saline (PBS) and Trypsin-EDTA (25-052-CI) obtained from Mediatech.
• PBS calcium and magnesium-free Phosphate Buffered Saline
• Trypsin-EDTA (25-052-CI) obtained from Mediatech.
• Cells were imaged on a MDS 1X5000 fluorescent microscope equipped with a 10X S Fluor objective, a Xenon light source, Chroma Filters for Dapi and Texas Red and a CCD camera. Hardware components were connected to a PC using Win2000 operating system and controlled with MetaX software (MDS Molecular Devices, Sunnyvale, CA USA). Images were captured and analyzed in 16-bit format using segmentation and morphological routines contained in the CellProfiler image analysis software (Broad
• PAD Plating density in 96-well plates
• Gin Glutamine (xlOOO cells/well)
• FBS Fetal Bovine Serum
• Results for the nine images per well were summed for each well and then averaged across duplicates.
• EC 50 S were generated by fitting the cell counts to a variable slope four parameter sigmoidal dose response curve using non-linear least squares method with the Trust-Region algorithm. Error bars on dose response points reflect standard deviations. Data was fit using five different sets of parameters forcing the top or the bottom to negative or positive (or zero) controls (Taxol 20nM, Etoposide 5uM) included on each plate. The different parameter sets were bottom to zero, bottom to the most potent of either of the positive controls, bottom to the positive control and the top to the negative control, bottom to zero and the top to the negative control and one curve was fit by letting the top and bottom float. All fits allowed the slope to float. The best fit for each compound was assessed by manual inspection after taking into account the quality of the fit and the biological relevance of the fit result. Inactive compounds were designated 9999 for EC 50 .
• the tumor cell line H1299 from ATCC was cultured in media with 10% Fetal Bovine Serum and without Pen-Strep.
• Cells were plated using a Multidrop on 96-well Corning plates at a density of 18K per ml and allowed to grow for 24 hours prior to the addition of compounds.
• Cell plating and further experiments were performed in RPMI with 5% FBS and 1% P/S.
• Compound dilutions were performed in DMSO on a Beckman FX equipped with a Span-8 pod.
• Training sets were drawn from treatment wells at concentrations in relation to the EC50 for that compound.
• Fig. 7 shows images from three wells of a six-point dose response with one well below the estimated EC50 and two above.
• One treatment training set is drawn from the first concentration above the EC50 and a different one from the second step.
• Classifiers generated to these training sets were applied to their respective concentrations. This strategy allowed for comparison of compounds with different EC50 response ranges to be compared.
• Control (DMSO) treated cells were drawn randomly from negative control wells contained in each plate. Training sets were generally 1000 cells, if available. As illustrated by Fig. 6, classifiers were made more robust by adding training sets of DMSO controls from other plates and compounds having a different mode of action.
• Fig. 8 shows how classifiers are trained and how new data is classified and analyzed to provide a prediction value.
• the process most likely responsible for a well's labeled results could be found by taking the ratio of the probability that the process is trial given the observed results to the probability the process is control.
• This is the likelihood ratio that can be written in the form of Bayes Theorem. Given a classifer's performance characteristics and number of cells classified as either trial or control, the binomial coefficient was used to find the likelihood ratio. This value is the prediction value reported in the figures which typically ranged from -2000 to +500, across many classifiers, cell lines and a few thousand compounds, with negative values predicting similarity to control and positive values meaning similarity to trial.
• Classifiers were generated against the 140 features for each cell. Ranges for features collected were standardized using the mean and standard deviation of each feature.
• Parameters were selected by examining the 'grid' of possible parameter values for a SVM using a radial kernel. C and gamma were varied between 0.01 and 10, and 0.001 and 1, respectively. Five steps were chosen in each direction to give 25 possible parameter pairs. For each parameter pair 3-fold cross-validation accuracy was calculated. SVM were trained using the full 'control' and 'experimental' training sets and then tested on the standardized 'control' and 'experimental' test sets, and basic quality control measures including the per- well false positive and negative rate were calculated. The composition of control training sets was varied to increase accuracy as described below.
• Training sets were created from treated wells spanning -1000 96- well plates screened over a year and a half.
• Compounds with known mechanisms of action were identified and used to investigate the ability of classifiers to generalize beyond the specific treatment for a given training.
• These groups included inhibitors to AuroraB, the 26S proteosome, Tubulin, Actin, Topoisomerase I, and antibiotic DNA synthesis.
• Classifiers created for these groups used training sets containing 40% each DMSO from the same plate as the trial and 40% from any other plate, in addition to 20% from wells treated with compounds that were not of similar MOA.
• R769 classifier control training sets contained cells treated with MG132, Taxol, Latrunculin A, and Camptothecin (and other similar compounds), but not any "AS" compounds, which had known AuroraB activity. After creating these classifiers, 100 compounds including the above mentioned, along with known inactive compounds and other controls, were retested in the PAD_48hr assay.
• a binary classifier is the tool used at the cell level to say whether the cell was more likely to have been generated via trial or control. As classifiers were trained performance characteristics were obtained. Given these characteristics and the fraction of a given well labeled as trial or control by the classifier the process most likely responsible for a well's labeled results could be found by taking the ratio of:
• a and b are the probability of given classifier labeling a cell trial or control, respectively, and x and y are the number of cells classified trial or control.
• x and y are the number of cells classified trial or control.
• svm classifiers are made to the moa control set of compounds.
• the number of experiments (essentially plates) the training set examples were drawn from is N.
• the classifier validation statistics are listed (True Pos, etc.) and the precision and recall of each svm against a test set of 100 compounds is listed as well. This is done for both the +1 set of svm' s and the +2 set.
• the assay was performed weekly with all plates plated with cell, dosed, fixed and stained as a group. Compounds were dosed 5 to a 96 well plate in 6-point dose response in duplicate for each 96 well plate. On each plate a negative control of DMSO was dosed at the same percentage of concentration as compound dosing and positive controls Taxol and Etoposide were dosed at a single concentration. These controls were used to calculate Z prime factors for each plate and to assess the staining intensity from the cell cycle plots.
• Digital images of DAPI stained nuclei were captured and segmented to locate nuclei and measure features such as intensity, area, shape and texture for each nuclei.
• As compounds were tested sample images from each concentration of the dose responses were visually inspected and a morphological category was assigned if appropriate. Dose response curves were inspected to assure proper EC50 determination and quality control assessments such as incorrect concentration range of dosing were annotated.
• Individual cell data, experimental properties and manual inspection results were stored in a custom built software system that allowed us to retrieve individual cell data by experimental properties such as compound name, concentration, and cell line and to filter out data from experiments of insufficient quality.
• Figs. 11 illustrates exemplary results from a proliferation-apoptosis-DNA content (PAD) assay
• Fig. 12 provides a graph of results obtained for inactive compounds screened by the subject method. Likelihood scores for true negatives and false positives are indicated.
• FIG. 13 is a table showing recall and precision results for selected classifiers.
• a threshold may be chosen for both the recall and precision in order to increase or decrease classifier robustness.
• FIG. 14 shows graphs illustrating the average recall performance of some classifiers, whereas FIG. 15 shows graphs illustrating the average precision

Landscapes

• Engineering & Computer Science (AREA)
• Chemical & Material Sciences (AREA)
• Crystallography & Structural Chemistry (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Health & Medical Sciences (AREA)
• Databases & Information Systems (AREA)
• Evolutionary Computation (AREA)
• General Health & Medical Sciences (AREA)
• Medical Informatics (AREA)
• Artificial Intelligence (AREA)
• Software Systems (AREA)
• Data Mining & Analysis (AREA)
• Life Sciences & Earth Sciences (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Bioinformatics & Computational Biology (AREA)
• Computing Systems (AREA)
• Theoretical Computer Science (AREA)
• Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
• Investigating Or Analysing Biological Materials (AREA)
• Image Processing (AREA)

Priority Applications (3)

Applications Claiming Priority (2)

Publications (1)

Family

ID=44258251

Family Applications (1)

Country Status (5)

Cited By (1)

Families Citing this family (9)

Citations (5)

Family Cites Families (12)

• 2011
  • 2011-01-05 CA CA2785835A patent/CA2785835C/en active Active
  • 2011-01-05 WO PCT/US2011/020262 patent/WO2011087945A1/en not_active Ceased
  • 2011-01-05 EP EP11733206.4A patent/EP2524337B1/en active Active
  • 2011-01-05 US US12/985,237 patent/US10311214B2/en active Active
  • 2011-01-05 JP JP2012548101A patent/JP5738894B2/ja not_active Expired - Fee Related

Patent Citations (5)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events