Mining and visualizing large anticancer drug discovery databases

Synergy between machine learning and natural products cheminformatics: Application to the lead discovery of anthraquinone derivatives

Cheminformatics utilizing machine learning (ML) techniques have opened up a new horizon in drug discovery. This is owing to vast chemical space expansion with rocketing numbers of expected hits and lead compounds that match druggable macromolecular targets, in particular from natural compounds. Due to the natural products' (NP) structural complexity, uniqueness, and diversity, they could occupy a bigger space in pharmaceuticals, allowing the industry to pursue more selective leads in the nanomolar range of binding affinity. ML is an essential part of each step of the drug design pipeline, such as target prediction, compound library preparation, and lead optimization. Notably, molecular mechanic and dynamic simulations, induced docking, and free energy perturbations are essential in predicting best binding poses, binding free energy values, and molecular mechanics force fields. Those applications have leveraged from artificial intelligence (AI), which decreases the computational costs required for such costly simulations. This review aimed to describe chemical space and compound libraries related to NPs. High-throughput screening utilized for fractionating NPs and high-throughput virtual screening and their strategies, and significance, are reviewed. Particular emphasis was given to AI approaches, ML tools, algorithms, and techniques, especially in drug discovery of macrocyclic compounds and approaches in computer-aided and ML-based drug discovery. Anthraquinone derivatives were discussed as a source of new lead compounds that can be developed using ML tools for diverse medicinal uses such as cancer, infectious diseases, and metabolic disorders. Furthermore, the power of principal component analysis in understanding relevant protein conformations, and molecular modeling of protein-ligand interaction were also presented. Apart from being a concise reference for cheminformatics, this review is a useful text to understand the application of ML-based algorithms to molecular dynamics simulation and in silico absorption, distribution, metabolism, excretion, and toxicity prediction.

Machine learning & deep learning in data-driven decision making of drug discovery and challenges in high-quality data acquisition in the pharmaceutical industry

Predicting novel small molecule bioactivities for the target deconvolution, hit-to-lead optimization in drug discovery research, requires molecular representation. Previous reports have demonstrated that machine learning (ML) and deep learning (DL) have substantial implications in virtual screening, peptide synthesis, drug ADMET screening and biomarker discovery. These strategies can increase the positive outcomes in the drug discovery process without false-positive rates and can be achieved in a cost-effective way with a minimum duration of time by high-quality data acquisition. This review substantially discusses the recent updates in AI tools as cheminformatics application in medicinal chemistry for the data-driven decision making of drug discovery and challenges in high-quality data acquisition in the pharmaceutical industry while improving small-molecule bioactivities and properties.

Accelerated structure-based design of chemically diverse allosteric modulators of a muscarinic G protein-coupled receptor

Design of ligands that provide receptor selectivity has emerged as a new paradigm for drug discovery of G protein-coupled receptors, and may, for certain families of receptors, only be achieved via identification of chemically diverse allosteric modulators. Here, the extracellular vestibule of the M2 muscarinic acetylcholine receptor (mAChR) is targeted for structure-based design of allosteric modulators. Accelerated molecular dynamics (aMD) simulations were performed to construct structural ensembles that account for the receptor flexibility. Compounds obtained from the National Cancer Institute (NCI) were docked to the receptor ensembles. Retrospective docking of known ligands showed that combining aMD simulations with Glide induced fit docking (IFD) provided much-improved enrichment factors, compared with the Glide virtual screening workflow. Glide IFD was thus applied in receptor ensemble docking, and 38 top-ranked NCI compounds were selected for experimental testing. In [(3)H]N-methylscopolamine radioligand dissociation assays, approximately half of the 38 lead compounds altered the radioligand dissociation rate, a hallmark of allosteric behavior. In further competition binding experiments, we identified 12 compounds with affinity of ≤30 μM. With final functional experiments on six selected compounds, we confirmed four of them as new negative allosteric modulators (NAMs) and one as positive allosteric modulator of agonist-mediated response at the M2 mAChR. Two of the NAMs showed subtype selectivity without significant effect at the M1 and M3 mAChRs. This study demonstrates an unprecedented successful structure-based approach to identify chemically diverse and selective GPCR allosteric modulators with outstanding potential for further structure-activity relationship studies.

Identification of novel radiosensitizers in a high-throughput, cell-based screen for DSB repair inhibitors

Most cancer therapies involve a component of treatment that inflicts DNA damage in tumor cells, such as double-strand breaks (DSBs), which are considered the most serious threat to genomic integrity. Complex systems have evolved to repair these lesions, and successful DSB repair is essential for tumor cell survival after exposure to ionizing radiation (IR) and other DNA-damaging agents. As such, inhibition of DNA repair is a potentially efficacious strategy for chemo- and radiosensitization. Homologous recombination (HR) and nonhomologous end-joining (NHEJ) represent the two major pathways by which DSBs are repaired in mammalian cells. Here, we report the design and execution of a high-throughput, cell-based small molecule screen for novel DSB repair inhibitors. We miniaturized our recently developed dual NHEJ and HR reporter system into a 384-well plate-based format and interrogated a diverse library of 20,000 compounds for molecules that selectively modulate NHEJ and HR repair in tumor cells. We identified a collection of novel hits that potently inhibit DSB repair, and we have validated their functional activity in a comprehensive panel of orthogonal secondary assays. A selection of these inhibitors was found to radiosensitize cancer cell lines in vitro, which suggests that they may be useful as novel chemo- and radio sensitizers. Surprisingly, we identified several FDA-approved drugs, including the calcium channel blocker mibefradil dihydrochloride, that demonstrated activity as DSB repair inhibitors and radiosensitizers. These findings suggest the possibility for repurposing them as tumor cell radiosensitizers in the future. Accordingly, we recently initiated a phase I clinical trial testing mibefradil as a glioma radiosensitizer.

Comprehensive Review in Current Developments of Benzimidazole-Based Medicinal Chemistry

The properties of benzimidazole and its derivatives have been studied over more than one hundred years. Benzimidazole derivatives are useful intermediates/subunits for the development of molecules of pharmaceutical or biological interest. Substituted benzimidazole derivatives have found applications in diverse therapeutic areas such as antiulcer, anticancer agents, and anthelmintic species to name just a few. This work systematically gives a comprehensive review in current developments of benzimidazole-based compounds in the whole range of medicinal chemistry as anticancer, antibacterial, antifungal, anti-inflammatory, analgesic agents, anti-HIV, antioxidant, anticonvulsant, antitubercular, antidiabetic, antileishmanial, antihistaminic, antimalarial agents, and other medicinal agents. This review will further be helpful for the researcher on the basis of substitution pattern around the nucleus with an aim to help medicinal chemists for developing an SAR on benzimidazole drugs/compounds.

Computational Search for Possible Mechanisms of 4-Thiazolidinones Anticancer Activity: The Power of Visualization

Public databases of NCI-60 tumor cell line screen results and measurements of molecular targets in the NCI-60 panel give the opportunity to assign possible anticancer mechanism to compounds with positive outcome from antitumor assay. Here, the novel protocol of NCI databases mining where inferences are based on the visualization is presented and utilized with the aim to identify putative biological routes of 4-thiazolidinones anticancer effect. As a result, highly potent 4-thiazolidinone-pyrazoline-isatin conjugates show the similarity of activity patterns with puromycin and CBU-028 and their pattern is also highly correlated with fraction of methylated CpG sites in CD34, AF5q31 and SYK. Several compounds from this group show strong negative correlation with fraction of methylated CpG sites in HOXA5. Thiopyrano[2,3-d][1,3]thiazol-2-ones bearing naphtoquinone fragment were found to possess the same activity pattern as fusarubin does. But none of the studied 4-thiazolidinone derivatives has activity fingerprint similar to standard anticancer agents. The obtained results bring medicinal chemistry closer to the understanding of basic nature of 4-thiazolidinones effect on cancer cells.

Machine learning prediction of cancer cell sensitivity to drugs based on genomic and chemical properties

Predicting the response of a specific cancer to a therapy is a major goal in modern oncology that should ultimately lead to a personalised treatment. High-throughput screenings of potentially active compounds against a panel of genomically heterogeneous cancer cell lines have unveiled multiple relationships between genomic alterations and drug responses. Various computational approaches have been proposed to predict sensitivity based on genomic features, while others have used the chemical properties of the drugs to ascertain their effect. In an effort to integrate these complementary approaches, we developed machine learning models to predict the response of cancer cell lines to drug treatment, quantified through IC₅₀ values, based on both the genomic features of the cell lines and the chemical properties of the considered drugs. Models predicted IC₅₀ values in a 8-fold cross-validation and an independent blind test with coefficient of determination R² of 0.72 and 0.64 respectively. Furthermore, models were able to predict with comparable accuracy (R² of 0.61) IC50s of cell lines from a tissue not used in the training stage. Our in silico models can be used to optimise the experimental design of drug-cell screenings by estimating a large proportion of missing IC₅₀ values rather than experimentally measuring them. The implications of our results go beyond virtual drug screening design: potentially thousands of drugs could be probed in silico to systematically test their potential efficacy as anti-tumour agents based on their structure, thus providing a computational framework to identify new drug repositioning opportunities as well as ultimately be useful for personalized medicine by linking the genomic traits of patients to drug sensitivity.

miRConnect: identifying effector genes of miRNAs and miRNA families in cancer cells

micro(mi)RNAs are small non-coding RNAs that negatively regulate expression of most mRNAs. They are powerful regulators of various differentiation stages, and the expression of genes that either negatively or positively correlate with expressed miRNAs is expected to hold information on the biological state of the cell and, hence, of the function of the expressed miRNAs. We have compared the large amount of available gene array data on the steady state system of the NCI60 cell lines to two different data sets containing information on the expression of 583 individual miRNAs. In addition, we have generated custom data sets containing expression information of 54 miRNA families sharing the same seed match. We have developed a novel strategy for correlating miRNAs with individual genes based on a summed Pearson Correlation Coefficient (sPCC) that mimics an in silico titration experiment. By focusing on the genes that correlate with the expression of miRNAs without necessarily being direct targets of miRNAs, we have clustered miRNAs into different functional groups. This has resulted in the identification of three novel miRNAs that are linked to the epithelial-to-mesenchymal transition (EMT) in addition to the known EMT regulators of the miR-200 miRNA family. In addition, an analysis of gene signatures associated with EMT, c-MYC activity, and ribosomal protein gene expression allowed us to assign different activities to each of the functional clusters of miRNAs. All correlation data are available via a web interface that allows investigators to identify genes whose expression correlates with the expression of single miRNAs or entire miRNA families. miRConnect.org will aid in identifying pathways regulated by miRNAs without requiring specific knowledge of miRNA targets.

Investigating the correlations among the chemical structures, bioactivity profiles and molecular targets of small molecules

MOTIVATION

Most of the previous data mining studies based on the NCI-60 dataset, due to its intrinsic cell-based nature, can hardly provide insights into the molecular targets for screened compounds. On the other hand, the abundant information of the compound-target associations in PubChem can offer extensive experimental evidence of molecular targets for tested compounds. Therefore, by taking advantages of the data from both public repositories, one may investigate the correlations between the bioactivity profiles of small molecules from the NCI-60 dataset (cellular level) and their patterns of interactions with relevant protein targets from PubChem (molecular level) simultaneously.

RESULTS

We investigated a set of 37 small molecules by providing links among their bioactivity profiles, protein targets and chemical structures. Hierarchical clustering of compounds was carried out based on their bioactivity profiles. We found that compounds were clustered into groups with similar mode of actions, which strongly correlated with chemical structures. Furthermore, we observed that compounds similar in bioactivity profiles also shared similar patterns of interactions with relevant protein targets, especially when chemical structures were related. The current work presents a new strategy for combining and data mining the NCI-60 dataset and PubChem. This analysis shows that bioactivity profile comparison can provide insights into the mode of actions at the molecular level, thus will facilitate the knowledge-based discovery of novel compounds with desired pharmacological properties.

AVAILABILITY

The bioactivity profiling data and the target annotation information are publicly available in the PubChem BioAssay database (ftp://ftp.ncbi.nlm.nih.gov/pubchem/Bioassay/).

Chemogenomic analysis identifies Macbecin II as a compound specific for SMAD4-negative colon cancer cells

The tumor suppressor gene, SMAD4, is mutated in approximately 30% of colon cancers. To identify compounds with enhanced potency on cells with a SMAD4-negative context, we combined genomic and cheminformatic analyses of publicly available data relating to the colon cancer cell lines within the NCI60 panel. Two groups of cell lines were identified with either wild-type or negative SMAD4 status. A cheminformatic analysis of the NCI60 screening data was carried out, which led to the identification of 14 compounds that preferentially inhibited cell growth of the SMAD4-negative cell lines. Using cell viability assays, the effect of these compounds was validated on four colon cancer cell lines: HCT-116 and HCT-15 (SMAD4-expressing), and HT-29 and COLO-205 (SMAD4-negative). Our data identified Macbecin II, a hydroquinone ansamycin antibiotic, as having increased potency in the SMAD4-negative cells compared to SMAD4 wild-type cells. In addition, we showed that silencing of SMAD4 using siRNA in HCT-116 enhanced Macbecin II potency. Our results demonstrate that Macbecin II is specifically active in colon cancer cells having a SMAD4-negative background and thus is a potential candidate for further investigation in a drug discovery perspective.

Mechanisms of multidrug resistance in cancer

The development of multidrug resistance (MDR) to chemotherapy remains a major challenge in the treatment of cancer. Resistance exists against every effective anticancer drug and can develop by numerous mechanisms including decreased drug uptake, increased drug efflux, activation of detoxifying systems, activation of DNA repair mechanisms, evasion of drug-induced apoptosis, etc. In the first part of this chapter, we briefly summarize the current knowledge on individual cellular mechanisms responsible for MDR, with a special emphasis on ATP-binding cassette transporters, perhaps the main theme of this textbook. Although extensive work has been done to characterize MDR mechanisms in vitro, the translation of this knowledge to the clinic has not been crowned with success. Therefore, identifying genes and mechanisms critical to the development of MDR in vivo and establishing a reliable method for analyzing clinical samples could help to predict the development of resistance and lead to treatments designed to circumvent it. Our thoughts about translational research needed to achieve significant progress in the understanding of this complex phenomenon are therefore discussed in a third section. The pleotropic response of cancer cells to chemotherapy is summarized in a concluding diagram.

Multiple-docking and affinity fingerprint methods for protein classification and inhibitors selection

The function-based protein classification holds tremendous promise for molecular recognition and the structure-based design process. We describe here a new strategy combined with multiple-docking tools and "affinity fingerprint" analysis technology to detect functional relationships among proteins based on the substrate binding features and protein-ligand interaction matrix and apply it successfully for the family of phospholipase A2 to investigate protein-ligand binding, function-based protein classification, and inhibitor selection, evaluation. Binding data and matrix were generated by multiple versus multiple-docking among 12 PLA2s and 84 PLA2 inhibitors. Three kinds of statistic techniques, principal component analysis, multidimensional scaling, and cluster algorithms, were chosen to distinguish the groups with similar binding characteristics. The 12 PLA2s were automatically categorized into reasonable subfamilies on the basis of the protein-ligand binding matrix, and the classifying problem of cPLA2 (PDB ID: 1CJY ) with relatively low homology was successfully dealt with. This approach was also used to identify and group the selective inhibitors against human nonpancreatic sPLA2. A sound pharmacophore has been defined from these selective inhibitors. It shows that the method is quite robust against individual data deviation, especially false positive, which makes it possible to be used in virtual screening with large enzyme families to generate selective inhibitors of targets on the basis of limited structural/function information.

Structural similarity assessment for drug sensitivity prediction in cancer

Background

The ability to predict drug sensitivity in cancer is one of the exciting promises of pharmacogenomic research. Several groups have demonstrated the ability to predict drug sensitivity by integrating chemo-sensitivity data and associated gene expression measurements from large anti-cancer drug screens such as NCI-60. The general approach is based on comparing gene expression measurements from sensitive and resistant cancer cell lines and deriving drug sensitivity profiles consisting of lists of genes whose expression is predictive of response to a drug. Importantly, it has been shown that such profiles are generic and can be applied to cancer cell lines that are not part of the anti-cancer screen. However, one limitation is that the profiles can not be generated for untested drugs (i.e., drugs that are not part of an anti-cancer drug screen). In this work, we propose using an existing drug sensitivity profile for drug A as a substitute for an untested drug B given high structural similarities between drugs A and B.

Results

We first show that structural similarity between pairs of compounds in the NCI-60 dataset highly correlates with the similarity between their activities across the cancer cell lines. This result shows that structurally similar drugs can be expected to have a similar effect on cancer cell lines. We next set out to test our hypothesis that we can use existing drug sensitivity profiles as substitute profiles for untested drugs. In a cross-validation experiment, we found that the use of substitute profiles is possible without a significant loss of prediction accuracy if the substitute profile was generated from a compound with high structural similarity to the untested compound.

Conclusion

Anti-cancer drug screens are a valuable resource for generating omics-based drug sensitivity profiles. We show that it is possible to extend the usefulness of existing screens to untested drugs by deriving substitute sensitivity profiles from structurally similar drugs part of the screen.

Discovery of drug-like inhibitors of an essential RNA-editing ligase in Trypanosoma brucei

Trypanosomatid RNA editing is a unique process and essential for these organisms. It therefore represents a drug target for a group of protozoa that includes the causative agents for African sleeping sickness and other devastating tropical and subtropical diseases. Here, we present drug-like inhibitors of a key enzyme in the editing machinery, RNA-editing ligase 1 (REL1). These inhibitors were identified through a strategy employing molecular dynamics to account for protein flexibility. A virtual screen of the REL1 crystal structure against the National Cancer Institute Diversity Set was performed by using AutoDock4. The top 30 compounds, predicted to interact with REL1's ATP-binding pocket, were further refined by using the relaxed complex scheme (RCS), which redocks the compounds to receptor structures extracted from an explicitly solvated molecular dynamics trajectory. The resulting reordering of the ligands and filtering based on drug-like properties resulted in an initial recommended set of 8 ligands, 2 of which exhibited micromolar activity against REL1. A subsequent hierarchical similarity search with the most active compound over the full National Cancer Institute database and RCS rescoring resulted in an additional set of 6 ligands, 2 of which were confirmed as REL1 inhibitors with IC(50) values of approximately 1 microM. Tests of the 3 most promising compounds against the most closely related bacteriophage T4 RNA ligase 2, as well as against human DNA ligase IIIbeta, indicated a considerable degree of selectivity for RNA ligases. These compounds are promising scaffolds for future drug design and discovery efforts against these important pathogens.

Perturbational profiling of nanomaterial biologic activity

Our understanding of the biologic effects (including toxicity) of nanomaterials is incomplete. In vivo animal studies remain the gold standard; however, widespread testing remains impractical, and the development of in vitro assays that correlate with in vivo activity has proven challenging. Here, we demonstrate the feasibility of analyzing in vitro nanomaterial activity in a generalizable, systematic fashion. We assessed nanoparticle effects in a multidimensional manner, using multiple cell types and multiple assays that reflect different aspects of cellular physiology. Hierarchical clustering of these data identifies nanomaterials with similar patterns of biologic activity across a broad sampling of cellular contexts, as opposed to extrapolating from results of a single in vitro assay. We show that this approach yields robust and detailed structure-activity relationships. Furthermore, a subset of nanoparticles were tested in mice, and nanoparticles with similar activity profiles in vitro exert similar effects on monocyte number in vivo. These data suggest a strategy of multidimensional characterization of nanomaterials in vitro that can inform the design of novel nanomaterials and guide studies of in vivo activity.

Design of fructose-2,6-bisphosphatase inhibitors: A novel virtual screening approach

Fructose-2,6-bisphosphatase (FBPase-2) is a switch between gluconeogenesis and glycolysis in the hepatic cells. The structural features required for inhibitory activity of FBPase-2 were unidentified; no leads are available for inhibiting this important enzyme. In this paper pharmacophore mapping, molecular docking methods were employed in a virtual screening strategy to identify leads for FBPase-2. A receptor based pharmacophore map was modeled which comprised of important interactions as observed in co-crystal of rat liver isozyme with the product inhibitor fructose-6-phosphate. The pharmacophore model was validated against two databases of best docked structural analogues of fructose-2,6-bisphosphate and fructose-6-phosphate. The query generated was submitted for flexible search of ligands in chemical databases, namely LeadQuest, Maybridge and NCI. The hits obtained were further screened by molecular docking using FlexX.

Response projected clustering for direct association with physiological and clinical response data

Background

Microarray gene expression data are often analyzed together with corresponding physiological response and clinical metadata of biological subjects, e.g. patients' residual tumor sizes after chemotherapy or glucose levels at various stages of diabetic patients. Current clustering analysis cannot directly incorporate such quantitative metadata into the clustering heatmap of gene expression. It will be quite useful if these clinical response data can be effectively summarized in the high-dimensional clustering display so that important groups of genes can be intuitively discovered with different degrees of relevance to target disease phenotypes.

Results

We introduced a novel clustering analysis approach, response projected clustering (RPC), which uses a high-dimensional geometrical projection of response data to the gene expression space. The projected response vector, which becomes the origin in the projected space, is then clustered together with the projected gene vectors based on their different degrees of association with the response vector. A bootstrap-counting based RPC analysis is also performed to evaluate statistical tightness of identified gene clusters. Our RPC analysis was applied to the in vitro growth-inhibition and microarray profiling data on the NCI-60 cancer cell lines and the microarray gene expression study of macrophage differentiation in atherogenesis. These RPC applications enabled us to identify many known and novel gene factors and their potential pathway associations which are highly relevant to the drug's chemosensitivity activities and atherogenesis.

Conclusion

We have shown that RPC can effectively discover gene networks with different degrees of association with clinical metadata. Performed on each gene's response projected vector based on its degree of association with the response data, RPC effectively summarizes individual genes' association with metadata as well as their own expression patterns. Thus, RPC greatly enhances the utility of clustering analysis on investigating high-dimensional microarray gene expression data with quantitative metadata.

A strategy for predicting the chemosensitivity of human cancers and its application to drug discovery

The U.S. National Cancer Institute has used a panel of 60 diverse human cancer cell lines (the NCI-60) to screen >100,000 chemical compounds for anticancer activity. However, not all important cancer types are included in the panel, nor are drug responses of the panel predictive of clinical efficacy in patients. We asked, therefore, whether it would be possible to extrapolate from that rich database (or analogous ones from other drug screens) to predict activity in cell types not included or, for that matter, clinical responses in patients with tumors. We address that challenge by developing and applying an algorithm we term "coexpression extrapolation" (COXEN). COXEN uses expression microarray data as a Rosetta Stone for translating from drug activities in the NCI-60 to drug activities in any other cell panel or set of clinical tumors. Here, we show that COXEN can accurately predict drug sensitivity of bladder cancer cell lines and clinical responses of breast cancer patients treated with commonly used chemotherapeutic drugs. Furthermore, we used COXEN for in silico screening of 45,545 compounds and identify an agent with activity against human bladder cancer.

Cancer cell-based genomic and small molecule screens

This chapter focuses on the promising post-genomic technologies being used for discovery of new, safer, and better cancer drugs and drug targets. Since cancer is largely a disease of the cell, usually involving unrestricted cell proliferation as a result of heritable genetic changes such as mutation, this chapter will focus on cell-centric technologies and their utility in addressing major questions in cancer biology. Recent advances in cell-based technology, including phenotypic assays, image-based readouts, primary tumor cell growth and maintenance in vitro, gene and small molecule delivery tools, and automated systems for cell manipulation, provide a novel means to understand the etiology and mechanisms of cancer as never before. In addition to the abundant tool sophistication, many aspects of cancer can be emulated and monitored in cell systems, which makes them ideal vehicles for exploitation to discover new targets and drugs. This chapter will first handle nomenclature and provide a context for a "good drug target" within the framework of the human genome, then overview functional genomic gene-based library screening approaches with specific applications to cancer target discovery. Second, small molecule screening applications will be handled, with an emphasis on the new paradigm of massively parallel screening and resultant multidimensional dataset analysis approaches to identify drug candidates, assign mechanism of action, and address problems in deriving selective and safe chemical entities.

Systems cell biology based on high-content screening

A new discipline of biology has emerged since 2004, which we call "systems cell biology" (SCB). Systems cell biology is the study of the living cell, the basic unit of life, an integrated and interacting network of genes, proteins, and myriad metabolic reactions that give rise to function. SCB takes advantage of high-content screening platforms, but delivers more detailed profiles of cellular systemic function, including the application of advanced reagents and informatics tools to sophisticated cellular models. Therefore, an SCB profile is a cellular systemic response as measured by a panel of reagents that quantify a specific set of biomarkers.

Small molecules, big players: the National Cancer Institute's Initiative for Chemical Genetics

In 2002, the National Cancer Institute created the Initiative for Chemical Genetics (ICG), to enable public research using small molecules to accelerate the discovery of cancer-relevant small-molecule probes. The ICG is a public-access research facility consisting of a tightly integrated team of synthetic and analytical chemists, assay developers, high-throughput screening and automation engineers, computational scientists, and software developers. The ICG seeks to facilitate the cross-fertilization of synthetic chemistry and cancer biology by creating a research environment in which new scientific collaborations are possible. To date, the ICG has interacted with 76 biology laboratories from 39 institutions and more than a dozen organic synthetic chemistry laboratories around the country and in Canada. All chemistry and screening data are deposited into the ChemBank web site (http://chembank.broad.harvard.edu/) and are available to the entire research community within a year of generation. ChemBank is both a data repository and a data analysis environment, facilitating the exploration of chemical and biological information across many different assays and small molecules. This report outlines how the ICG functions, how researchers can take advantage of its screening, chemistry and informatic capabilities, and provides a brief summary of some of the many important research findings.

Correlations between the activities of 19 anti-tumor agents and the intracellular glutathione concentrations in a panel of 14 human cancer cell lines: comparisons with the National Cancer Institute data

The aim of this work was 2-fold: (i) to identify correlations between the activities of pairs of 19 anti-tumor agents in a mini-panel of 14 human cancer cell lines of diverse origins with the goal of validating the panel, and (ii) to look for correlations between the activities of 19 standard anti-tumor agents and the intracellular concentrations of glutathione (GSH). Validation with analogous data from the National Cancer Institute (NCI) Developmental Therapeutics Program was made. The cell growth inhibition potencies of the anti-tumor agents [cisplatin, carboplatin, oxaliplatin, DACH-Pt, melphalan, chlorambucil, thiotepa, busulfan, doxorubicin, etoposide, camptothecin, vinblastine, podophyllotoxin, colchicine, taxol, hydroxyurea, methotrexate, 5-azacytidine and 5-fluorouracil (5-FU)] were estimated in 14 cancer cell lines by their GI50 values. An enzymatic assay based on the method of Tietze was employed to measure intracellular total GSH concentrations. The Delta method was used to compare pairs of anti-tumor agents; similarities and differences in activity profiles (mean graphs) were evaluated by regression analysis. Most, but not all, of the correlations could be explained based on similarities in the mechanisms of action and many correlations/non-correlations were also observed in the NCI data. Some correlations were unexpected however, and not seen in the NCI data. For example, strong positive correlations (P < 0.01) were found between the GI50 values of melphalan/chlorambucil and the anti-mitotic agents. Similarly unexpected, a strong positive correlation was observed between methotrexate and cisplatin (P < 0.01). Interestingly, moderate to strong negative correlations (P < 0.01-0.05) were found between the GI50 values of 5-FU and the anti-mitotic agents/melphalan/chlorambucil. Significant positive correlations between intracellular GSH concentrations and GI50 values were found only for thiotepa (P < 0.05) and doxorubicin (P < 0.01). Data from a NCI panel of 34 cancer cell lines showed no correlations between GSH levels and the GI50 values of the same 19 compounds. In conclusion, a panel of 14 human cancer cell lines of diverse origin was used to identify similarities and differences in the activities of standard anti-tumor agents. The level of significance was stronger with the 34 cell lines of the NCI, however. Our results indicate that GSH intracellular concentrations correlate with resistance only with doxorubicin and thiotepa in these cell lines.

Data mining of NCI's anticancer screening database reveals mitochondrial complex I inhibitors cytotoxic to leukemia cell lines

Mitochondria are principal mediators of apoptosis and thus can be considered molecular targets for new chemotherapeutic agents in the treatment of cancer. Inhibitors of mitochondrial complex I of the electron transport chain have been shown to induce apoptosis and exhibit antitumor activity. In an effort to find novel complex I inhibitors which exhibited anticancer activity in the NCI's tumor cell line screen, we examined organized tumor cytotoxicity screening data available as SOM (self-organized maps) (http://www.spheroid.ncifcrf.gov) at the developmental therapeutics program (DTP) of the National Cancer Institute (NCI). Our analysis focused on an SOM cluster comprised of compounds which included a number of known mitochondrial complex I (NADH:CoQ oxidoreductase) inhibitors. From these clusters 10 compounds whose mechanism of action was unknown were tested for inhibition of complex I activity in bovine heart sub-mitochondrial particles (SMP) resulting in the discovery that 5 of the 10 compounds demonstrated significant inhibition with IC50's in the nM range for three of the five. Examination of screening profiles of the five inhibitors toward the NCI's tumor cell lines revealed that they were cytotoxic to the leukemia subpanel (particularly K562 cells). Oxygen consumption experiments with permeabilized K562 cells revealed that the five most active compounds inhibited complex I activity in these cells in the same rank order and similar potency as determined with bovine heart SMP. Our findings thus fortify the appeal of mitochondrial complex I as a possible anticancer molecular target and provide a data mining strategy for selecting candidate inhibitors for further testing.

Multi-target strategies for the improved treatment of depressive states: Conceptual foundations and neuronal substrates, drug discovery and therapeutic application

Major depression is a debilitating and recurrent disorder with a substantial lifetime risk and a high social cost. Depressed patients generally display co-morbid symptoms, and depression frequently accompanies other serious disorders. Currently available drugs display limited efficacy and a pronounced delay to onset of action, and all provoke distressing side effects. Cloning of the human genome has fuelled expectations that symptomatic treatment may soon become more rapid and effective, and that depressive states may ultimately be "prevented" or "cured". In pursuing these objectives, in particular for genome-derived, non-monoaminergic targets, "specificity" of drug actions is often emphasized. That is, priority is afforded to agents that interact exclusively with a single site hypothesized as critically involved in the pathogenesis and/or control of depression. Certain highly selective drugs may prove effective, and they remain indispensable in the experimental (and clinical) evaluation of the significance of novel mechanisms. However, by analogy to other multifactorial disorders, "multi-target" agents may be better adapted to the improved treatment of depressive states. Support for this contention is garnered from a broad palette of observations, ranging from mechanisms of action of adjunctive drug combinations and electroconvulsive therapy to "network theory" analysis of the etiology and management of depressive states. The review also outlines opportunities to be exploited, and challenges to be addressed, in the discovery and characterization of drugs recognizing multiple targets. Finally, a diversity of multi-target strategies is proposed for the more efficacious and rapid control of core and co-morbid symptoms of depression, together with improved tolerance relative to currently available agents.

Linking tumor cell cytotoxicity to mechanism of drug action: an integrated analysis of gene expression, small-molecule screening and structural databases

An integrated, bioinformatic analysis of three databases comprising tumor-cell-based small molecule screening data, gene expression measurements, and PDB (Protein Data Bank) ligand-target structures has been developed for probing mechanism of drug action (MOA). Clustering analysis of GI50 profiles for the NCI's database of compounds screened across a panel of tumor cells (NCI60) was used to select a subset of unique cytotoxic responses for about 4000 small molecules. Drug-gene-PDB relationships for this test set were examined by correlative analysis of cytotoxic response and differential gene expression profiles within the NCI60 and structural comparisons with known ligand-target crystallographic complexes. A survey of molecular features within these compounds finds thirteen conserved Compound Classes, each class exhibiting chemical features important for interactions with a variety of biological targets. Protein targets for an additional twelve Compound Classes could be directly assigned using drug-protein interactions observed in the crystallographic database. Results from the analysis of constitutive gene expressions established a clear connection between chemo-resistance and overexpression of gene families associated with the extracellular matrix, cytoskeletal organization, and xenobiotic metabolism. Conversely, chemo-sensitivity implicated overexpression of gene families involved in homeostatic functions of nucleic acid repair, aryl hydrocarbon metabolism, heat shock response, proteasome degradation and apoptosis. Correlations between chemo-responsiveness and differential gene expressions identified chemotypes with nonselective (i.e., many) molecular targets from those likely to have selective (i.e., few) molecular targets. Applications of data mining strategies that jointly utilize tumor cell screening, genomic, and structural data are presented for hypotheses generation and identifying novel anticancer candidates.

Transcriptional profiling of sorghum induced by methyl jasmonate, salicylic acid, and aminocyclopropane carboxylic acid reveals cooperative regulation and novel gene responses

We have conducted a large-scale study of gene expression in the C4 monocot sorghum (Sorghum bicolor) L. Moench cv BTx623 in response to the signaling compounds salicylic acid (SA), methyl jasmonate (MeJA), and the ethylene precursor aminocyclopropane carboxylic acid. Expression profiles were generated from seedling root and shoot tissue at 3 and 27 h, using a microarray containing 12,982 nonredundant elements. Data from 102 slides and quantitative reverse transcription-PCR data on mRNA abundance from 171 genes were collected and analyzed and are here made publicly available. Numerous gene clusters were identified in which expression was correlated with particular signaling compound and tissue combinations. Many genes previously implicated in defense responded to the treatments, including numerous pathogenesis-related genes and most members of the phenylpropanoid pathway, and several other genes that may represent novel activities or pathways. Genes of the octadecanoic acid pathway of jasmonic acid (JA) synthesis were induced by SA as well as by MeJA. The resulting hypothesis that increased SA could lead to increased endogenous JA production was confirmed by measurement of JA content. Comparison of responses to SA, MeJA, and combined SA+MeJA revealed patterns of one-way and mutual antagonisms, as well as synergistic effects on regulation of some genes. These experiments thus help further define the transcriptional results of cross talk between the SA and JA pathways and suggest that a subset of genes coregulated by SA and JA may comprise a uniquely evolved sector of plant signaling responsive cascades.

Diagnostic plots for detecting outlying slides in a cDNA microarray experiment

Different sources of systematic and random error variations are often observed in cDNA microarray experiments. A simple scatter plot is commonly used to examine outlying slides that have unusual expression patterns or larger variability than other slides. These outlying slides tend to have large impacts on the subsequent analyses, such as identification of differentially expressed genes and clustering analysis. However, it is difficult to select outlying slides rigorously and consistently based on subjective human pattern recognition on their scatter plots. A graphical method and a rigorous diagnostic measure are proposed to detect outlying slides. The proposed graphical method is easy to implement and shown to be quite effective in detecting outlying slides in real microarray data sets. This diagnostic measure is also informative to compare variability among slides. Two cDNA microarray data sets are carefully examined to illustrate the proposed approach. A 3840-gene microarray experiment for neuronal differentiation of cortical stem cells and a 2076-gene microarray experiment for anticancer compound time-course expression of the NCI-60 cancer cell lines.

Workshop on Cancer Biometrics: Identifying Biomarkers and Surrogates of Cancer in Patients

The current excitement about molecular targeted therapies has driven much of the recent dialog in cancer diagnosis and treatment. Particularly in the biologic therapy of cancer, identifiable antigenic T-cell targets restricted by MHC molecules and the related novel stress molecules such as MICA/B and Letal allow a degree of precision previously unknown in cancer therapy. We have previously held workshops on immunologic monitoring and angiogenesis monitoring. This workshop was designed to discuss the state of the art in identification of biomarkers and surrogates of tumor in patients with cancer, with particular emphasis on assays within the blood and tumor. We distinguish this from immunologic monitoring in the sense that it is primarily a measure of the tumor burden as opposed to the immune response to it. Recommendations for intensive investigation and targeted funding to enable such strategies were developed in seven areas: genomic analysis; detection of molecular markers in peripheral blood and lymph node by tumor capture and RT-PCR; serum, plasma, and tumor proteomics; immune polymorphisms; high content screening using flow and imaging cytometry; immunohistochemistry and tissue microarrays; and assessment of immune infiltrate and necrosis in tumors. Concrete recommendations for current application and enabling further development in cancer biometrics are summarized. This will allow a more informed, rapid, and accurate assessment of novel cancer therapies.

Multi-class cancer classification by total principal component regression (TPCR) using microarray gene expression data

DNA microarray technology provides a promising approach to the diagnosis and prognosis of tumors on a genome-wide scale by monitoring the expression levels of thousands of genes simultaneously. One problem arising from the use of microarray data is the difficulty to analyze the high-dimensional gene expression data, typically with thousands of variables (genes) and much fewer observations (samples), in which severe collinearity is often observed. This makes it difficult to apply directly the classical statistical methods to investigate microarray data. In this paper, total principal component regression (TPCR) was proposed to classify human tumors by extracting the latent variable structure underlying microarray data from the augmented subspace of both independent variables and dependent variables. One of the salient features of our method is that it takes into account not only the latent variable structure but also the errors in the microarray gene expression profiles (independent variables). The prediction performance of TPCR was evaluated by both leave-one-out and leave-half-out cross-validation using four well-known microarray datasets. The stabilities and reliabilities of the classification models were further assessed by re-randomization and permutation studies. A fast kernel algorithm was applied to decrease the computation time dramatically. (MATLAB source code is available upon request.).

Induction of lysosomal membrane permeabilization by compounds that activate p53-independent apoptosis

The p53 protein activates cellular death programs through multiple pathways. Because the high frequency of p53 mutations in human tumors is believed to contribute to resistance to commonly used chemotherapeutic agents, it is important to identify drugs that induce p53-independent cell death and to define the mechanisms of action of such drugs. Here we screened a drug library (the National Cancer Institute mechanistic set; 879 compounds with diverse mechanisms of actions) and identified 175 compounds that induced caspase cleavage of cytokeratin-18 in cultured HCT116 colon cancer cells at <5 microM. Interestingly, whereas most compounds elicited a stronger apoptotic response in cells with functional p53, significant apoptosis was observed also in p53-null cells. A subset of 15 compounds showing weak or no dependence on p53 for induction of apoptosis was examined in detail. Of these compounds, 11 were capable of activating caspase-3 in enucleated cells. Seven such compounds with nonnuclear targets were found to induce lysosomal membrane permeabilization (LMP). Translocation of the lysosomal proteases cathepsin B and cathepsin D into the cytosol was observed after treatment with these drugs, and apoptosis was inhibited by pepstatin A, an inhibitor of cathepsin D. Apoptosis depended on Bax, suggesting that LMP induced a mitochondrial apoptotic pathway. We conclude that a large number of potential anticancer drugs induce p53-independent apoptosis and that LMP is a mediator of many such responses.

Structure-based discovery of human l-xylulose reductase inhibitors from database screening and molecular docking

Human L-xylulose reductase (XR) is an enzyme of the glucuronic acid/uronate cycle of glucose metabolism and is a possible target for treatment of the long-term complications of diabetes. In this study we utilised the molecular modelling program DOCK to analyse the 249,071 compounds of the National Cancer Institute Database and retrieved those compounds with high predicted affinity for XR. Several carboxylic acid-based compounds were tested and shown to inhibit XR. These included nicotinic acid (IC50=100 microM), benzoic acid (IC50=29 microM) and their derivatives. These results extend and improve upon the activities of known, commercially available inhibitors of XR such as the aliphatic fatty acid n-butyric acid (IC50=64 microM). To optimise the interaction between the inhibitor and the holoenzyme, the program GRID was used to design de novo compounds based on the inhibitor benzoic acid. The inclusion of a hydroxy-phenyl group and a phosphate to the benzoic acid molecule increased the net binding energy by 1.3- and 2.4-fold, respectively. The resultant compounds may produce inhibitors with improved specificity for XR.

A Flexible Data Analysis Tool for Chemical Genetic Screens

High-throughput assays generate immense quantities of data that require sophisticated data analysis tools. We have created a freely available software tool, SLIMS (Small Laboratory Information Management System), for chemical genetics which facilitates the collection and analysis of large-scale chemical screening data. Compound structures, physical locations, and raw data can be loaded into SLIMS. Raw data from high-throughput assays are normalized using flexible analysis protocols, and systematic spatial errors are automatically identified and corrected. Various computational analyses are performed on tested compounds, and dilution-series data are processed using standard or user-defined algorithms. Finally, published literature associated with active compounds is automatically retrieved from Medline and processed to yield potential mechanisms of actions. SLIMS provides a framework for analyzing high-throughput assay data both as a laboratory information management system and as a platform for experimental analysis.

Subchronic exposure to TCDD, PeCDF, PCB126, and PCB153: effect on hepatic gene expression

We employed DNA microarray to identify unique hepatic gene expression patterns associated with subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and other halogenated aromatic hydrocarbons (HAHs). Female Harlan Sprague-Dawley rats were exposed for 13 weeks to toxicologically equivalent doses of four different HAHs based on the toxic equivalency factor of each chemical: TCDD (100 ng/kg/day), 2,3,4,7,8-pentachlorodibenzofuran (PeCDF; 200 ng/kg/day), 3,3',4,4',5-pentachlorobiphenyl (PCB126; 1,000 ng/kg/day), or 2,2',4,4',5,5'-hexachlorobiphenyl (PCB153; 1,000 microg/kg/day). Global gene expression profiles for each exposure, which account for 8,799 gene probe sets contained on Affymetrix RGU34A GeneChips, were compared by principal components analysis. The aryl hydrocarbon receptor (AhR) ligands TCDD, PeCDF, and PCB126 produced very similar global gene expression profiles that were unique from the nonAhR ligand PCB153, underscoring the extensive impact of AhR activation and/or the resulting hepatic injury on global gene expression in female rat liver. Many genes were co-expressed during the 13-week TCDD, PeCDF, or PCB126 exposures, including classical AhR-regulated genes and some genes not previously characterized as being AhR regulated, such as carcinoembryonic-cell adhesion molecule 4 (C-CAM4) and adenylate cyclase-associated protein 2 (CAP2). Real-time reverse-transcriptase polymerase chain reaction confirmed the increased expression of these genes in TCDD-, PeCDF-, and PCB126-exposed rats as well as the up- or down-regulation of several other novel dioxin-responsive genes. In summary, DNA microarray successfully identified dioxin-responsive genes expressed after exposure to AhR ligands (TCDD, PeCDF, PCB126) but not after exposure to the non-AhR ligand PCB153. Together, these findings may help to elucidate some of the fundamental features of dioxin toxicity and may further clarify the biologic role of the AhR signaling pathway.

Applications of machine learning and high-dimensional visualization in cancer detection, diagnosis, and management

Recent technical advances in combinatorial chemistry, genomics, and proteomics have made available large databases of biological and chemical information that have the potential to dramatically improve our understanding of cancer biology at the molecular level. Such an understanding of cancer biology could have a substantial impact on how we detect, diagnose, and manage cancer cases in the clinical setting. One of the biggest challenges facing clinical oncologists is how to extract clinically useful knowledge from the overwhelming amount of raw molecular data that are currently available. In this paper, we discuss how the exploratory data analysis techniques of machine learning and high-dimensional visualization can be applied to extract clinically useful knowledge from a heterogeneous assortment of molecular data. After an introductory overview of machine learning and visualization techniques, we describe two proprietary algorithms (PURS and RadViz) that we have found to be useful in the exploratory analysis of large biological data sets. We next illustrate, by way of three examples, the applicability of these techniques to cancer detection, diagnosis, and management using three very different types of molecular data. We first discuss the use of our exploratory analysis techniques on proteomic mass spectroscopy data for the detection of ovarian cancer. Next, we discuss the diagnostic use of these techniques on gene expression data to differentiate between squamous and adenocarcinoma of the lung. Finally, we illustrate the use of such techniques in selecting from a database of chemical compounds those most effective in managing patients with melanoma versus leukemia.

Classification of a large anticancer data set by Adaptive Fuzzy Partition

An Adaptive Fuzzy Partition (AFP) algorithm, derived from Fuzzy Logic concepts, was used to classify an anticancer data set, including about 1300 compounds subdivided into eight mechanisms of action. AFP classification builds relationships between molecular descriptors and bio-activities by dynamically dividing the descriptor hyperspace into a set of fuzzy subspaces. These subspaces are described by simple linguistic rules, from which scores ranging between 0 and 1 can be derived. The latter values define, for each compound, the degrees of membership of the different mechanisms analyzed. A particular attention was devoted to develop structure-activity relations that have a real utility. Then, well-defined and widely accepted protocols were used to validate the models by defining their robustness and prediction ability. More particularly, after selecting the most relevant descriptors with help of a genetic algorithm, a training set of 640 compounds was isolated by a rational procedure based on Self-Organizing Maps. The related AFP model was then validated with help of a validation set and, above all, of cross-validation and Y-randomization procedures. Good validation scores of about 80% were obtained, underlining the robustness of the model. Moreover, the prediction ability was evaluated with 374 test compounds that had not been used to establish the model and 77% of them were predicted correctly.

Biological mechanism profiling using an annotated compound library

We present a method for testing many biological mechanisms in cellular assays using an annotated library of 2036 small organic molecules. This annotated compound library represents a large-scale collection of compounds with diverse, experimentally confirmed biological mechanisms and effects. We found that this chemical library is (1) more structurally diverse than conventional, commercially available libraries, (2) enriched in active compounds in a tumor cell viability assay, and (3) capable of generating hypotheses regarding biological mechanisms underlying cellular processes. We elucidated biological mechanisms relevant to the antiproliferative activity of 85 compounds from this library that were selected using a high-throughput cell viability screen. We developed a novel automated scoring system for identifying statistically enriched mechanisms among such a subset of compounds. This scoring system can identify both previously known and potentially novel antiproliferative mechanisms.

Transcriptomic analysis of the NCI-60 cancer cell lines

Pharmacogenomics aims at molecular subsetting of patients for more effective therapy. Transcriptomic profiling of the 60 human cancer cell lines (the NCI-60) used by the US National Cancer Institute serves that aim because the cells have been treated with > 100,000 chemical compounds over the last 13 years. Patterns of potency can be mapped into molecular structures of the compounds or into molecular characteristics of the cells. We discuss conceptual and experimental aspects of the profiling, as well as a number of bioinformatic computer programs that we have developed for biological interpretation of the profiles.

Web-based tools for mining the NCI databases for anticancer drug discovery

In this paper, we describe the development of a set of integrated Web-based tools for mining the National Cancer Institute's (NCI) anticancer databases for anticancer drug discovery. For data mining, three different correlation algorithms were implemented, which included the commonly used Pearson's correlation algorithm available from the NCI's COMPARE program, the Spearman's and Kendall's correlation algorithms. In addition, we implemented the p-value test to evaluate the significance of the correlation results. These Web-based data mining tools allow robust analysis of the correlation between the in vitro anticancer activity of the drugs in the NCI anticancer database, the protein levels and mRNA levels of molecular targets (genes) in the NCI 60 human cancer cell lines for identification of potential lead compounds for a specific molecular target and for study of the molecular mechanism action of a drug. Examples were provided to identify PKC ligands using a lead compound and to identify potential ErbB-2 inhibitors using the mRNA levels of ErbB-2 in the NCI 60 tumor cell lines.

Novel technologies for virtual screening

There are several methods for virtual screening of databases of small organic compounds to find tight binders to a given protein target. Recent reviews in Drug Discovery Today have concentrated on screening by docking and by pharmacophore searching. Here, we complement these reviews by focusing on virtual screening methods that are based on analyzing ligand similarity on a structural level. Specifically, we concentrate on methods that exploit structural properties of the complete ligand molecules, as opposed to using just partial structural templates, such as pharmacophores. The in silico procedure of virtual screening (VS) and its relationship to the experimental procedure, HTS, is discussed, new developments in the field are summarized and perspectives on future research are offered.

Cytostatic and cytotoxic activity of synthetic genistein glycosides against human cancer cell lines

Genistein, the principal soy isoflavone, is a molecule of great interest as an innovative chemotherapeutic agent or as a lead-compound in anticancer drug design. To enhance intrinsic activity of genistein and to explore its pharmacophoric potential, its glycosidic derivatives were synthesized. On the basis of structural features and calculated lipophilicity coefficient (ClogP) the derivatives were classified as hydrophilic (i.e. those containing free sugar moiety) or lipophilic (i.e. those with alkylated or acylated sugar hydroxyls). The in vitro cytostatic and cytotoxic studies showed hydrophilic glycosides to be practically inactive against human cancer cell lines when compared to the free aglycone. On the contrary, lipophilic glycosides were significantly more active than the parent isoflavone although the correlation between ClogP and the activity was not clear. On the basis of GI50 and LC50 values two of the most active glycosides were found to be several times more potent in their cytostatic and cytotoxic effect than genistein. Additionally all lipophilic glycosides were revealed to exhibit different mode of action in comparison to genistein. It may suggest that these compounds do not undergo rapid biodegradation, either in culture media or inside cells, and exert their biological effects primarily as intact molecules.

Comparing cDNA and oligonucleotide array data: concordance of gene expression across platforms for the NCI-60 cancer cells

Microarray gene-expression profiles are generally validated one gene at a time by real-time RT-PCR. A different approach is described, based on simultaneous mutual validation of large numbers of genes using two different expression-profiling platforms.

Microarray gene-expression profiles are generally validated one gene at a time by real-time RT-PCR. We describe here a different approach based on simultaneous mutual validation of large numbers of genes using two different expression-profiling platforms. The result described here for the NCI-60 cancer cell lines is a consensus set of genes that give similar profiles on spotted cDNA arrays and Affymetrix oligonucleotide chips. Global concordance is parameterized by a 'correlation of correlations' coefficient.

Data Mining the NCI Cancer Cell Line Compound GI50Values: Identifying Quinone Subtypes Effective Against Melanoma and Leukemia Cell Classes

Using data mining techniques, we have studied a subset (1400) of compounds from the large public National Cancer Institute (NCI) compounds data repository. We first carried out a functional class identity assignment for the 60 NCI cancer testing cell lines via hierarchical clustering of gene expression data. Comprised of nine clinical tissue types, the 60 cell lines were placed into six classes-melanoma, leukemia, renal, lung, and colorectal, and the sixth class was comprised of mixed tissue cell lines not found in any of the other five classes. We then carried out supervised machine learning, using the GI(50) values tested on a panel of 60 NCI cancer cell lines. For separate 3-class and 2-class problem clustering, we successfully carried out clear cell line class separation at high stringency, p < 0.01 (Bonferroni corrected t-statistic), using feature reduction clustering algorithms embedded in RadViz, an integrated high dimensional analytic and visualization tool. We started with the 1400 compound GI(50) values as input and selected only those compounds, or features, significant in carrying out the classification. With this approach, we identified two small sets of compounds that were most effective in carrying out complete class separation of the melanoma, non-melanoma classes and leukemia, non-leukemia classes. To validate these results, we showed that these two compound sets' GI(50) values were highly accurate classifiers using five standard analytical algorithms. One compound set was most effective against the melanoma class cell lines (14 compounds), and the other set was most effective against the leukemia class cell lines (30 compounds). The two compound classes were both significantly enriched in two different types of substituted p-quinones. The melanoma cell line class of 14 compounds was comprised of 11 compounds that were internal substituted p-quinones, and the leukemia cell line class of 30 compounds was comprised of 6 compounds that were external substituted p-quinones. Attempts to subclassify melanoma or leukemia cell lines based upon their clinical cancer subtype met with limited success. For example, using GI(50) values for the 30 compounds we identified as effective against all leukemia cell lines, we could subclassify acute lymphoblastic leukemia (ALL) origin cell lines from non-ALL leukemia origin cell lines without significant overlap from non-leukemia cell lines. Based upon clustering using GI(50) values for the 60 cancer cell lines laid out by the RadViz algorithm, these two compound subsets did not overlap with clusters containing any of the NCI's 92 compounds of known mechanism of action, a few of which are quinones. Given their structural patterns, the two p-quinone subtypes we identified would clearly be expected to possess different redox potentials/substrate specificities for enzymatic reduction in vivo. These two p-quinone subtypes represent valuable information that may be used in the elucidation of pharmacophores for the design of compounds to treat these two cancer tissue types in the clinic.