Online informatics resources to facilitate cancer target and chemical probe discovery

This review highlights the cutting-edge informatics resources available to explore cancer genomics, biological, and chemical space to facilitate target and therapeutic discovery in cancer.

Abstract

The advances in cancer genomics, chemical biology, high-throughput screening technologies, and synthetic medicinal chemistry have tremendously expanded the biological space of cancer targets and chemical space of bioactive small molecules to interrogate oncogenic signaling. To explore and leverage these exponentially growing cancer-associated data, a great number of computational tools, databases, and algorithms have been developed. This review summarizes recent cancer-related web resources that allow researchers working at the interface of chemical, biological, and cancer genomics fields to integrate clinical and genomics data for specific actionable targets and selective chemical compounds to facilitate cancer therapeutic discovery.

1. Introduction

Cancer is the second leading cause of death worldwide. Based on the World Health Organization (WHO) records, nearly 10 million people died from cancer in 2018. Tremendous effort has been made over the past decades to define the landscape of the cancer genome, understand the molecular mechanisms of the disease, and develop new clinical strategies for therapeutic intervention in cancer. Large-scale cancer genomics initiatives such as The Cancer Genome Atlas (TCGA)1 and Therapeutically Applicable Research to Generate Effective Treatments (TARGET)2 Programs of the National Cancer Institute (NCI) and the International Cancer Genomics Consortium (ICGC)3,4 have enabled a systematic discovery of previously unrecognized oncogenic pathways and individual proteins as new promising targets for therapeutic intervention in cancer.5–7 CRISPR-Cas9 and RNAi loss-of-function screens combined with the profiling of cancer cell line sensitivity to clinically relevant chemical compounds revealed new mechanisms of tumorigenesis and uncovered novel targets for therapeutic intervention.8–11 Simultaneously, the advances in synthetic chemistry, high-throughput screening technology, and assay development significantly expanded the chemical space of low molecular weight compounds to interrogate oncogenic signaling.12–16 The breakthroughs in experimental cancer genomics, proteomics, chemical biology, and synthetic chemistry demand the development of new cutting-edge computational tools to explore and integrate these multidimensional data to facilitate cancer therapeutic discovery.17,18

In this review, we highlight informatics resources available for the multidisciplinary community to facilitate the discovery of new promising therapeutic targets and high-quality chemical probes to interrogate oncogenic signaling. We specifically emphasize web-servers with a well-developed user interface that allows the exploration and analysis of large-scale datasets without requiring special knowledge in bioinformatics or programming skills (Table 1). First, we will provide an overview of web resources and databases focused on the identification of cancer-associated genes and genomic alterations. Then, we will overview web servers that are developed to identify new oncogenic pathways and target proteins based on genetics and chemical profiling of cancer cell line dependency, which will be followed by a discussion on recent advances in the development of computational tools to evaluate the chemical space, properties of bioactive compounds, and high-quality chemical probes to interrogate specific biological targets.

Table 1. The key online informatics resources for cancer target and chemical probe discovery.

Main focus	Resource	Summary	Data type & size
Repositories for cancer genomics data
Access to cancer patient genomics and clinical data	GDC	A repository for the large-scale cancer genomics data. Provides programmatic access and limited analytical tools to explore cancer patient genomic and clinical data	Genomics and clinical data for >83 700 samples from 15 large-scale cancer genomics programs
		https://gdc.cancer.gov
Access to cancer patient genomics and clinical data; cancer cell genomics	COSMIC	A curated database of mutations, DNA copy number, mRNA expression, gene-fusions or translations derived from cancer cell lines and patient samples. Includes association between mutations and drug resistance. Enables mapping of the point mutations on the protein structure	Genomics data for >1 391 300 samples from 466 cancer genomics studies. The data includes samples from cancer patients and cancer cell lines
		https://cancer.sanger.ac.uk/cosmic
Access to genomics and drug sensitivity of cancer cell lines	CCLE	The Cancer Cell Line Encyclopedia contains genomics data, including mRNA expression, DNA copy number, methylation, mutations, and fusion/translocation data, along with the sensitivity of cancer cells to drug treatment or individual gene knockout	Genomics and cell sensitivity data for >84 000 genes determined in 1457 cell lines
		https://portals.broadinstitute.org/ccle
Resources and analytical tools to explore cancer genome and proteome
Genomic alterations and their biological and clinical significance in cancer patients	cBioPortal	The well-developed web resource to explore, visualize, and analyze cancer genomics and clinical data generated in large-scale cancer genomic studies. Multiple analytical tools are available, including the distribution of the genomic alterations across multiple cancer types, analysis of mutual exclusivity, co-expression, and patient survival, and other. Well-integrated with multiple other online resources and databases	Genomics data for >2800 cell lines; genomics and clinical data for >82 000 patient samples from 280 cancer genomics studies
		https://www.cbioportal.org
Cancer driver mutations	Tumor portal	Tumor portal allows to explore the mutation frequency and functional impact of the mutations to uncover novel cancer driver genes	Mutation frequency and functional significance determined for >18 000 genes in 4742 human cancers across 21 tumor types
		http://www.tumorportal.org
Cancer proteome	The Cancer Proteome Atlas	The database of the reverse-phase protein array (RPPA) data derived cancer patient samples and cancer cell lines. The graphical interface and implemented analytical tools allows to determine the correlations between the protein level, gene expression, patient survival, and drug sensitivity	RPPA data derived from >8000 TCGA samples and >650 cancer cell lines
		https://tcpaportal.org/tcpa
Target discovery & mechanism of oncogenic signaling
Relationship between cancer cell sensitivity to chemical perturbagens and genomics alterations (mainly DNA copy number and mRNA expression alterations)	Cancer Therapeutics Response Portal	The CTRP enables the identification of correlations between the DNA copy number and mRNA expression alterations and the sensitivity of different cancer cell types to pharmacologically active compounds with known targets	Cell viability and genomics data derived from 860 cell lines treated by >480 compound
		https://portals.broadinstitute.org/ctrp.v2.1
Relationship between cancer cell sensitivity to chemical perturbagens and genomics alterations (mainly mutations)	Genomics of Drug Sensitivity in Cancer	The GDSC enables the identification of correlations between the gene mutations and the sensitivity of different cancer cell types to pharmacologically active compounds with known targets	Cell viability data determined in 988 cell lines treated by nearly 400 clinically relevant compounds. >446 000 dose–response curves; >570 000 genomic associations
		https://www.cancerrxgene.org
Relationship between cancer cell sensitivity to chemical perturbagens and genomics alterations. Comparison and correlation between different datasets	CellMinerCDB	Provides a user-friendly interface to access, explore, and compare the large datasets generated in multiple projects, including CTRP, GDSC, and other	Matches molecular features of >1000 cell lines and cell response to >20 000 compounds
		https://discover.nci.nih.gov/cellminercdb/
Cancer-associated protein–protein interaction networks	OncoPPi portal	Provide a user interface to explore the protein–protein interaction (PPI) networks experimentally determined for focused libraries of cancer-associated proteins, as well as the PPI-drug connectivity, and PPI essentiality for cancer cell viability. The PPI networks are integrated with multiple cancer-focused and general structural, biological, and clinical online resources and databases	The networks experimentally determined for >3600 PPIs in cancer cell lines
		http://oncoppi.emory.edu
Cancer cell dependency on specific proteins and drug treatment	Dependency Map	An interactive resource to facilitate the discovery of cancer cells dependency on specific proteins or drug treatment	Genomic alterations determined for >27 500 genes in >1700 cell lines; genetic dependency of >700 cell lines on >18 300 genes; sensitivity of >550 cell lines to >4500 compounds
		https://depmap.org
Chemical probe identification
Compound selectivity profiling	Probes and Drugs portal	The P&D portal provides detailed annotations for a large set of experimentally tested biologically active compounds to identify potent and selective chemical probes	Detailed annotations for biological activity and selectivity of >60 000 compounds, including >12 200 approved drugs
		https://www.probes-drugs.org
Identification of high-quality chemical probes	The Small Molecule Suite	The Small Molecule Suite aims to improve design of new compound collections for focused screening campaigns. SelectivitySelectR, SimilaritySelectR, and LibraryR are implemented to identify compounds with a high selectivity against a specific protein or group of target proteins and a significant chemical and phenotypical similarity	The reported activity of >1400 compounds experimentally determined for multiple target proteins
		https://labsyspharm.shinyapps.io/smallmoleculesuite/
Design of compound collections for focused screening campaigns	ChemicalProbes Portal	ChemicalProbes Portal provides manually curated annotations for high-quality chemical probes	The description of 188 compounds associated with 205 target proteins, including potency, selectivity, evidence for target engagement, and the mechanism of action
		https://www.chemicalprobes.org

2. Resources to identify cancer-associated proteins

2.1. COSMIC

Mutations are among the most frequent genomic alterations in cancer. The Catalogue of Somatic Mutations in Cancer (COSMIC)19 was the first database of systematically curated and annotated somatic mutations discovered in cancer cells. COSMIC has been launched in 2004 with the annotations of only four mutated oncogenes: HRAS, NRAS, KRAS, and BRAF.20 Now it provides the detailed information for nearly 6 000 000 coding mutations and more than 19 000 000 non-coding variants across 1 391 372 samples from 466 large-scale cancer genomics studies.19 This is the largest database of somatic mutations in cancer. Besides the mutations, COSMIC includes the annotations for 1 179 545 DNA copy-number, 9 147 833 mRNA expression, and 7 879 142 DNA methylation variants. Furthermore, the COSMIC-3D tool has been designed to enhance the understanding of the impact of mutations on the protein structure.21 Together, COSMIC is a major resource to explore the frequency and impact of somatic mutations in cancer.

2.2. cBioPortal

The cancer genomics programs, such as the TCGA launched in 2005 and ICGC launched in 2008, have allowed the comprehensive large-scale next-generation sequencing of the entire cancer genome and determined the landscape of genomic alterations in cancer. Consequently, multiple web-based servers and algorithms have been developed to enhance the availability, analysis, and integration of patient cancer genomics profiles with clinical and pharmacological data.22 Perhaps, one of the most widely used resources to explore, visualize, and analyze the genomics data derived from cancer patients is the cBioPortal.23,24 While cBioPortal was originally designed to facilitate the public access to the TCGA data, it now also provides access to the TARGET and ICGC studies as well as other curated projects. The data from 273 studies cover the DNA copy-number, mRNA and miRNA expression, protein-level and phosphoprotein level (RPPA), DNA methylation, and clinical data collected from more than 82 000 patients. Numerous analytical tools have been implemented in cBioPortal to facilitate the discovery of new oncogenic pathways and prioritization of proteins as potential therapeutic targets, including the analysis of mutual exclusivity of genomics alterations, mRNA co-expression, mutation, copy number, and mRNA expression enrichment analysis, and patient survival data. Through the integration with different other databases, including COSMIC,19 Cancer HotSpots,25,26 3D Hotspots,27 OncoKB,28 CIVic,29 and My Cancer Genome,30 cBioPortal provides detailed annotations for every mutant variant with the information about the variant recurrence, clinical actionability, and structural mapping. To illustrate the cBioPortal functionality, we used it to explore the landscape of genomic alterations in pancreatic cancer. Analysis of the pancreatic adenocarcinoma PanCancer Atlas dataset31 indicates frequent mutations in several major tumor driver genes, including KRAS (mutated in 65.4% patients), TP53 (59.8%), SMAD4 (20.7%), CDKN2A (19.6%). SMAD4 and CDKN2A, and CDKN2B are also frequently deleted in pancreatic cancer. While GATA6, MIB1, MYC, and NDRG1 are amplified in more than 10% of samples (Fig. 1A). The substitution of KRAS Gly12 located in KRAS P-loop by Asp, Val, or Arg are the most frequent KRAS mutations in pancreatic cancer (Fig. 1B), and they strongly correlate with decreased patient survival (Fig. 1C). Indeed, targeting KRAS signaling is an established though challenging strategy for therapeutic intervention in pancreatic cancer.32 Thus, the cBioPortal can clearly summarize and illustrate existing clinical and cancer genomics data, facilitating the utilization of these data for therapeutic target discovery.

Fig. 1. cBioPortal provides a framework to explore cancer patient genomics and clinical data. A) The OncoPrint panel shows frequent alterations of KRAS, TP53, SMAD4, CDKN2A, CDKN2B, GATA6, MIB1, MYC, and NDRG1 in patients with pancreatic cancer. Green, black, and brown bars indicate samples with missense, truncating, and inframe mutations, respectively. Purple bars indicate samples with fusions. Blue and red bars show samples with DNA amplification or deep deletion, respectively. Grey bars indicate patients with no alterations. B) The frequency and positions of KRAS mutants within the RAS domain are shown along with the location of KRAS G12D on the KRAS protein surface. C) cBioPortal allows to determine the correlation between KRAS G12D mutation status and decreased survival of pancreatic cancer patients.

2.3. Genomic Data Commons

The amount and complexity of the genomics data generated over the last decades raised an urgent need for data harmonization, curation, and standardization of data processing pipelines and annotations. To address this critical issue, the National Cancer Institute (NCI) has launched the Genomic Data Commons (GDC) project and the GDC Data Portal.33,34 While the TCGA samples have been originally aligned against the Genome Reference Consortium (GRC) hg18 or hg19 reference genomes, the GDC provides access to the “harmonized” genomics datasets re-aligned to the most recent Genome Reference Consortium build (hg38) using specifically developed GDC bioinformatics tools.35 Now GDC serves as the largest repository of cancer genomics data accessible programmatically through the established API. Furthermore, the GDC Data Portal provides a user-friendly interface to explore the frequency of mutations and copy-number alterations and their clinical impact determined in 13 different campaigns for more than 60 cancers. For example, a query of the GDC Data Portal for EGFR shows the distribution of EGFR alterations across different cancer types. EGFR is amplified in GBM, esophageal carcinoma, ovarian, lung, head and neck, and multiple other cancers (Fig. 2A), and it is mutated in more than 10% patients with glioblastoma multiform (GBM), lung adenocarcinoma, skin melanoma, and uterine carcinoma (Fig. 2B). EGFR A289V, G598V, and L858R are the most frequent EGFR mutations (Fig. 2C). While EGFR A289V and G598V appear in 6% of GBM patients, L858R is the most frequent EGFR mutation in lung cancer (18% samples). EGFR L858R mutation correlates with decreased survival of lung adenocarcinoma patients (p-value = 0.02, Fig. 2D) and inhibition of EGFR L858R is a promising therapeutic approach in lung cancer.36,37

Fig. 2. Analysis of EGFR alterations in different tumor types performed with the GDC Data Portal. A) Distribution of EGFR DNA copy-number alterations. Red bars show frequency of EGFR amplification, blue bars show the frequency of EGFR deletions. B) Distribution of EGFR mutations across different cancer types. C) A schematic diagram shows the location of EGFR mutated restudies in EGFR structural domains. The height of the lines reflects the mutation frequency. Blue, red, and magenta circles indicate missense, stop gained, and frameshift mutations, respectively. D) The survival curves generated by the GDC Data Portal show the correlation between EGFR L858R mutation and decreased survival of lung cancer patients.

2.4. TumorPortal

While cBioPortal and GDC Data Portal allow to explore and analyze the cancer genomics and clinical data for user-defined sets of genes, cancer types, and patient cohorts, several servers provide access to pre-processed datasets, including the GDAC Firehose (https://gdac.broadinstitute.org) and TumorPortal.38 The TumorPortal was developed based on the systematic analysis of somatic mutations in 4742 human cancers across 21 tumor types studied in the TCGA and non-TCGA projects.38 The TumorPortal provides the graphical maps for the mutations identified in more than 18 000 genes, along with the distribution of the mutation rates across different cancer types. To estimate the overall significance of the mutations, three different algorithms (MutSigCV, MutSigCL, and MutSigFN39,40) were used to evaluate the significance of mutation burden, clustering, and functional impact of the mutations. The combined single p-value and the false discovery rate adjusted q-value were derived as overall quantitative metrics for the mutation significance. The genes are further classified in three groups as highly significant mutated genes, significantly mutated, and genes near significance. Together, the TumorPortal helps to uncover highly compelling novel cancer driver genes for further exploration in detailed studies.

2.5. Cancer Proteome Atlas

The large-scale systematic profiling of DNA copy-number alterations and mRNA expression across different cancer types have revealed a relatively low correlation between the DNA or RNA level and the level of the protein coded by the corresponding genes.41,42 To address this critical issue and facilitate the discovery of proteins as actionable therapeutic targets, a quantitative antibody-based reverse-phase protein array (RPPA) technology has been developed.43 The application of the RPPA to characterize more than 8000 TCGA samples and more than 650 cancer cell lines has defined the landscape of the cancer proteome that is available to explore through The Cancer Proteome Atlas (TCPA).44 The TCPA web server includes the graphical interface to explore the cancer proteome based on the data source or specific cancer type and provide the tools to determine the correlations between the protein level, gene expression, patient survival, and drug sensitivity. For example, phosphorylation of AKT, a serine/threonine kinase, by PDK1 at Thr308 and by PDK2 or mTORC2 at Ser473 are among the major driving events in GBM.45 A query of the TCPA portal clearly indicates that the total protein level of AKT is about the average in GBM (Fig. 3A). In contrast, both AKT pT308 (Fig. 3B) and pS473 (Fig. 3C) are significantly elevated in GBM comparing to all other cancer types. The correlation analysis shows a strong correlation between AKT activation and phosphorylation of well-defined AKT substrates, including GSK3B pS9, TSC2 pT1462, and AKT1S1 pT246 (Fig. 3D–F). Thus, the TCPA data provides further support of established oncogenic pathways and may reveal new functional connectivity between cancer-associated proteins.

Fig. 3. The TCPA portal enables analysis of protein-level and phosphoprotein level (RPPA) data across 32 cancer types. A) The distribution of the total protein level of AKT across multiple cancer types. B) The distribution of the protein level of AKT phosphorylated at T308. C) The distribution of the protein level of AKT phosphorylated at S473. D) The protein level of AKT p308 correlates with the phosphorylation of GSK3B at S9. E) The protein level of AKT p308 correlates with the phosphorylation of TSC2 at T1462. F) The protein level of AKT p308 correlates with the phosphorylation of AKT1S1 at T246.

3. Cancer target discovery and prioritization

3.1. Dependency map

Unraveling of cancer human genome has established a defined set of cancer-associated proteins frequently altered in different cancer types. To determine the tumor dependency on these putative therapeutic targets and uncover the molecular mechanisms underlying their oncogenic potential, several genetic and chemical approaches have been developed. The large-scale RNAi and CRISPR loss-of-function screens allowed the discovery of hundreds of novel proteins that play essential roles in cancer cell survival and proliferation. These data, available through the Dependency Map (DepMap) portal,46 provide an invaluable resource to identify proteins that are essential for the viability of specific cancer cell types. For example, the association between decreased viability of skin cancer cell lines and BRAF knockout can be found using the DepMap Data Explorer tool (Fig. 4A). Furthermore, skin cancer cells show a strong co-dependency on BRAF knockout and knockout of its well-established downstream effectors, kinases ERK (MAPK1), MAP kinase phosphatase DUSP4, and transcription regulators SOX10 and MITF (Fig. 4B). Thus, in agreement with clinical data,47 DepMap data clearly indicate that inhibition of the BRAF pathway can provide a promising strategy to control the growth and survival of skin cancer cells.

Fig. 4. The DepMap Data Explorer allows the examination of cancer cell dependency on individual gene knockout. A) BRAF knockout significantly decreases the viability of skin cancer cell lines comparing to other cancer types. B) Skin cancer cells show a strong co-dependency on BRAF knockout and knockout of its well-established downstream effectors MAPK1, DUSP4, SOX10, and MITF.

3.2. Resources to explore cancer cell sensitivity to chemical perturbagens

Complementary to the genetics approach, tremendous efforts have been made to determine the sensitivity of cancer cells to the perturbation of oncogenic pathways by chemical compounds.48–50 Several publicly available resources, including the Cancer Therapeutics Response Portal (CTRP)51–53 and Genomics of Drug Sensitivity in Cancer (GDSC)54 have been developed providing the access and analytical tools to determine the connectivity between the cancer genomic background and response of cancer cells to pharmacologically active compounds with defined targets (Table 1). The GDSC portal allows the rapid analysis of the viability of 988 cell lines treated by nearly 400 clinically relevant compounds, including FDA-approved cancer drugs and compounds in clinical trials. The CTRP portal allows the analysis of sensitivity of the 860 cell lines to 481 compounds, including both, approved drugs or compounds in clinical trials, and other pharmacologically active compounds with known mechanism of action. Through the annotations of the cell lines with the gene mutations, DNA copy number alterations, and expression number both resources allow identification of new connectivity between the known compound targets and specific cancer genomic background. For example, a query of GDSC portal for BRAF inhibitors shows that the cell sensitivity data is available for an FDA-approved potent BRAF inhibitor dabrafenib.55 In agreement with the DepMap BRAF knockout data, skin melanoma cell lines show the highest sensitivity to dabrafenib in the GDSC dataset (Fig. 5A). BRAF is frequently mutated in melanoma. A comparative analysis of the sensitivity of melanoma cell lines with and without BRAF mutation demonstrates that the inhibitory effect of dabrafenib on the viability of BRAF mutant cell lines is significantly higher as compared with that of BRAF wild type cell lines (p < 0.001) (Fig. 5B).

Fig. 5. The GDSC, CTRP, and CellMinerCDB portals facilitate the analysis of cancer cell sensitivity to clinically relevant compounds. A) The bar graph shows the sensitivity of different cancer cell types. Skin melanoma cell lines show the highest sensitivity to the BRAF inhibitor dabrafenib. B) Skin melanoma cell lines harboring BRAF V600E mutation are more sensitive to dabrafenib comparing to cell lines with the wild-type BRAF. C) CellMinerCDB shows a significant correlation between expression of BRAF in skin cancer cell lines determined in GDSC and CTRP datasets. D) CellMinerCDB indicates a significant correlation between the activity (act) of dabrafenib as an inhibitor of skin cancer cell viability determined in GDSC and CTRP datasets.

Recently, to improve the integration, annotation, and reproducibility of the analysis of different pharmacological and cancer genomics datasets, the Integrative Cross-Database Genomics and Pharmacogenomics Database, CellMinerCDB has been developed.56 The CellMinerCDB provides a user-friendly interface to access, explore, and analyze the large datasets generated in multiple projects, including CTRP, GDSC, Cancer Cell Line Encyclopedia (CCLE),49 and NCI-60.57 By matching the molecular features of over 1000 cell lines and drug response measured in different assays for more than 20 000 compounds, the CellMinerCDB allows rapid identification of the most consistent and reproducible data across the datasets. For examples, CellMinerCDB shows a significant correlation between both, the expression of BRAF in skin cancer cell lines determined in GDSC and CTRP datasets (Pearson correlation R = 0.41, p-value = 0.02, Fig. 5C) and the sensitivity of skin cancer cell lines to dabrafenib (R = 0.59, p-value = 0.01, Fig. 5D).

3.3. OncoPPi Portal

The advances in cancer genomics and proteomics have revealed a critical role of genomic alterations in the acquisition of cancer hallmarks. The alterations in protein sequence or expression ultimately lead to a rewired network of protein–protein interactions (PPI), which control critical functions and physiological states of the cell. Discovery and understanding of cancer-associated PPIs would facilitate the development of new biological models and uncover new mechanisms of oncogenic signaling for therapeutic interrogation. Toward this goal, our team has developed a high-throughput screening platform to detect PPIs between cancer-associated proteins in the context of cancer cells that resulted in the network of oncogenic PPIs, termed the OncoPPi network version 1.58 To enable streamlined and integrated analysis of the PPI datasets, we have developed the OncoPPi Portal, a web-based resource that integrates the network of experimentally detected PPIs with cancer genomics, pharmacological and protein structural data.59,60 The OncoPPi Portal allows to access and explore a high-quality cancer-focused PPI network integrated with the analysis of mutual exclusivity of genomic alterations, cellular co-localization of interacting proteins, domain–domain interactions, and therapeutic connectivity. As an example, the exploration of the OncoPPi network has revealed multiple novel oncogenic PPIs, including new mechanisms for the regulation of major tumor driver MYC.16,58,61,62 To facilitate data mining for prioritization of cancer-associated PPIs and individual proteins for further experiments, every protein in the OncoPPi network is annotated and connected with a number of external general and cancer-focused resources, including the cBioPortal, TumorPortal, and the NCI Cancer Target Discovery and Development (CTD2) dashboard.63 Together, the OncoPPi Portal provides a powerful framework to explore cancer interactome and generate new hypotheses and biological models for cancer target discovery. A detailed description of the OncoPPi Portal functionality and examples of its application are provided elsewhere.59,60

4. Chemical probes and compound selectivity

The large-scale pharmacogenomics profiling integrated with the genetic RNAi and CRISPR-Cas9 approaches tremendously expanded our understanding of molecular mechanisms of cancer cell growth and survival. These efforts facilitated the discovery and development of new strategies to control tumorigenesis therapeutically. Meanwhile, several challenges in the phenotypic approach have been recognized, including the polypharmacology and missannotations of the small-molecule compounds.64 The urgent need for high-quality, well-validated and annotated chemical probes has emerged from frequently observed broad off-target effects of compounds included in the screening libraries, including the approved drugs.65,66 Several web-based servers and databases have been developed to address this critical issue (Table 1).

4.1. Probes and Drugs portal

The Probes and Drugs (P&D) portal has been launched in 2017 to facilitate the discovery of potent and selective chemical probes through the exploration of the bioactive compound space.67 Currently, the P&D portal provides detailed annotations for more than 60 000 unique compounds assembled from 61 public and commercial compound collections, including cancer-focused libraries such as the NIH Approved Oncology Drugs, NIH Mechanistic Set, and the Broad Institute Informer Set compounds characterized in a panel of 860 cancer cell lines. The P&D portal compound dataset includes a total of 12 291 approved drugs. Meanwhile, there is a growing demand for high-quality chemical probe compounds.65,68 The drugs are characterized mostly by bioavailability and safety and must be effective against certain medical conditions. However, the mechanism of action of a drug is not necessarily known and its target specificity can be limited. In contrast, the chemical probes are used to interrogate specific biological targets and processes, and thus must demonstrate a high potency and target specificity as well as a well-defined mechanism of action. Although the bioavailability is not a critical parameter for a chemical probe, a sufficient cell-permeability and low toxicity are required for cell-based and in vivo studies.

Moreover, P&D portal includes compounds defined as chemical probes in several other databases, including the Broad CTRP, ChemicalProbes Portal discussed below, sets from Structural Genomics Consortium (SGC, https://www.thesgc.org), the NIH Molecular Libraries Program (MLP), as well as compounds that are indicated as probes in Nature Chemical Biology collections (https://www.nature.com/nchembio/collections). Currently, P&D Portal classifies more than 4000 molecules as chemical probes. It should be noted that each of the source databases uses different criteria to define a molecule as a probe. Therefore, the quality of chemical probes available through the P&D portal can vary significantly. To facilitate the detailed exploration of compound properties, each compound in the P&D portal database is linked to different types of chemical, biological, and clinical data, and the powerful user interface allows rapid analysis and filtering of the compounds based on the desired properties. For example, the analysis of chemical probe molecules indicates the association of these compounds with 8899 target proteins of different classes, such as kinases, G protein-coupled receptors, ion channels, epigenetic regulators, and other. In turn, the association of target proteins with biological pathways allows to identify multiple cancer-associated pathways that can be regulated by the selected chemical probe compounds. That includes cell cycle, metabolism of proteins, cellular response to stress, regulation of immune and neuronal systems and other pathways. Furthermore, the user interface allows a user to easily identify compounds that are specifically associated with a particular target or pathway. For example, 888 out of ∼4000 probe compounds can regulate cell cycle through the kinase inhibition, and 89 of them are approved drugs. The further filtering of the compounds based on the number of reported targets indicates that 234 out of 888 compounds have five or more known targets, and only 73 inhibitors have been associated with only one specific kinase.

The P&D Portal also allows to search for compounds that are active against a specific target. For example, a search for BCL2 inhibitors reveals a total of 42 compounds. A further filtering for highly potent (potency >8) and selective (potency-selectivity score >0.8) BCL2 inhibitors revealed only one compound, venetoclax, the first approved BCL2 protein–protein interaction inhibitor69 (Fig. 6A). Through a search for compounds that share significant similarity with a user-provided structure, one can generate a set of structurally similar compounds and identify their potential targets. For example, a search for compounds that share at least 50% similarity with a BRD4 inhibitor JQ1 (ref. 70) results in 16 molecules, that can be further clustered together based on the similarity of physico-chemical properties, including the topological polar surface area (TPSA), clog P, number of H-bond donors and acceptors, molecular weight, the total number of rings, the number of aromatic rings and rotatable bonds, as well as the SMILES similarity (Fig. 6B). The target-association analysis shows that all JQ-1 analogs can inhibit bromodomain-containing proteins.

Fig. 6. The Probes and Drugs (P&D) portal and Small Molecule Suite provide detailed annotations for thousands of chemical compounds. A) The information available on the P&D portal for potent, selective, and FDA-approved BCL2 inhibitor venetoclax is shown as an example. B) The P&D portal allows the compound similarity search and cluster analysis based on different compound properties. The heatmap shows a cluster analysis performed for 16 analogs of BRD4 inhibitor JQ-1. Blue and red colors indicate the lowest and highest similarity, respectively. C) The SelectivitySelectR tool of the Small Molecule Suite links the protein targets with more than 1400 unique compounds with reported activity against multiple protein targets. The plot shows the correlation between the selectivity score and binding affinity (Kd) of identified CDK4 inhibitors.

4.2. The Small Molecule Suite

Recently, the Small Molecule Suite has been developed to facilitate and improve the design of new compound collections for focused screening campaigns.71 Three tools, SelectivitySelectR, SimilaritySelectR, and LibraryR are implemented to identify compounds that show a high selectivity against a target of interest, compounds with significant chemical and phenotypical similarity, and to design focused compound sets to target a specific group of proteins. Through the integration of PubChem and ChEMBL annotations, the SelectivitySelectR tool links the protein targets with more than 1400 unique compounds with reported activity against multiple protein targets. The detailed information on the binding selectivity, target coverage, induced cellular phenotypes, chemical structure, and stage of clinical development are provided for each compound. Several metrics are implemented to prioritize the most selective inhibitors, including the differential IC₅₀ values, defined as the difference between the on-target and off-target IC₅₀ values in the log10 scale divided by 3, and the Selectivity score72 that evaluates compound-target pairs based on the distributions of the on-target and off-target affinities. For example, a search for CDK4 inhibitors reveals a total of 66 compounds that can inhibit CDK4 with the IC₅₀ values of 8.2 nM to more than 10 μM (Fig. 6C). However, only three inhibitors demonstrate the differential IC₅₀ > 100 fold, on-target IC₅₀ < 100 nM, and the selectivity scores > 0.5: palbociclib (IC₅₀ = 8.2 nM), AT-7519 (IC₅₀ = 18 nM), and SNS-032 (IC₅₀ = 69 nM). These compounds are assigned the “most selective” type of CDK4 inhibitors. In addition, alvocidib (IC₅₀ = 100 nM) is indicated as the “semi-selective” inhibitor. The SelectivitySelectR shows that alvocidib is also a potent inhibitor of multiple cyclin dependent kinases, besides CDK4. Another powerful tool implemented in the Small Molecule Suite, the LibraryR, helps to facilitate the design of compound libraries focused on a specific group of proteins. For example, a simultaneous search for the most-selective inhibitors of all 13 CDKs with Kd ≤ 10 nM revealed a total of 15 compounds, including the FDA-approved drug palbociclib, and multiple compounds in clinical trials, such as alvocidib, ribociclib, abemaciclib, and other. Together, the Small Molecule Suite and Probes and Drugs portals provide powerful resources and convenient user interfaces to explore the chemical space of pharmacologically active compounds for chemical probe discovery.

4.3. ChemicalProbes Portal

The analysis of selectivity profiles of well-known potent and widely used compounds, including the approved drugs, indicates that the majority of these molecules have multiple well-defined protein targets. This polypharmacological or multi-target effect of the available inhibitors significantly limits their application as chemical probes to interrogate specific biological pathways, including the mechanisms of oncogenic signaling. The ChemicalProbe Portal65 has been specifically designed to facilitate the identification of high-quality chemical probes through the manual curation of publications by the Scientific Advisory Board (SAB) of medicinal chemistry and chemical biology experts. The current version of the portal contains descriptions for 188 compounds associated with 205 target proteins. Each compound has been reviewed by the SAB members. Based on the information and data quality available for the compound, including potency, selectivity, evidence for target engagement, and the understanding of the mechanism of action, each compound is rated using the star-rating system:

4 stars: recommended as a probe for this target.

3 stars: the best available probe for this target, or a high-quality probe that is a useful orthogonal tool.

2 stars: insufficient validation data to recommend.

1 star: not recommended as a probe for this target.

Currently, 42 out of 188 compounds are rated as 4-star probes and 81 out of 188 have a rating between 3 and 4. These 123 compounds are recommended for use as specific chemical probes to modulate the activity of the associated targets. Besides the overall rating, detailed information and SAB comments are provided for each compound in the ChemicalProbes database, including the chemical structure, PubChem and ChEMBL IDs, target proteins, mechanism of action, recommended concentration for use in cells, PAINs evaluation, compound potency determined in cells and model organisms, associated references, and availability of the compound from commercial sources. For example, a query of the ChemicalProbes Portal for ERBB2 inhibitors reveals two compounds: afatinib and CP-724714. Afatinib can also inhibit EGFR and thus it is not recommended as a chemical probe for ERBB2. In contrast, CP-724714 inhibits ERBB2 with the IC₅₀ = 10 nM in the cell-free assays and IC₅₀ = 32 nM in cell-based assays and demonstrates >640-fold selectivity over GFR, InsR, IRG-1R, PDGFR, VEGFR2, ABL, SRC, and c-MET, and >1000-fold selectivity over ABL, SRC, MET, JUN, JNK2, JNK3, ZAP-70, CDK2, and CDK5. CP-724714 is a cell-permeable PAINs free compound, and it is commercially available from at least 5 vendors. The recommended concentration for use in cells ranges from 50 nM to 10 μM. CP-724714 has a 4-star rating and it is recommended by SAB members as a chemical probe for ERBB2. This example illustrates the power and applicability of ChemicalProbe Portal for the rapid identification of high-quality chemical probes to interrogate specific biological targets.

5. Conclusions

Recent advances in large-scale high-throughput genomics, chemical biology, and medicinal chemistry have significantly expanded the biological and chemical space for cancer targets and therapeutic discovery. A great number of cutting-edge computational tools and databases have been developed to enable the exploration and leveraging of the multidimensional data by chemists, biologists, and clinicians. This review aims to provide a brief overview of the well-developed web servers which allow us to link clinical and cancer genomics data with specific actionable targets and validated potent and selective compounds. Multiple servers, including the COSMIC database, GDC Data Portal, cBioPortal, Tumor Portal, and Cancer Proteome Atlas provide rapid and user-friendly access to uncover new potential cancer-driving proteins and oncogenic pathways from cancer patient genomics data generated in large-scale projects such as TCGA, ICGC, and TARGET. The Dependency MAP, CellMinerCDB, CTRP, and GDCS servers are invaluable to discover new tumor dependency on a specific target protein, while the exploration of experimentally-determined cancer-specific protein–protein interaction networks available through the OncoPPi Portal can uncover new connectivity between the oncogenic pathways. Finally, the development of ChemicalProbes Portal, Probes and Drugs portal, and the Small Molecule Suite reflects the emerging need for high-quality well-annotated chemical probe development to interrogate oncogenic signaling. We believe that the review of these essential cutting-edge tools will be helpful for researchers working at the interfaces of chemical, biological, and cancer genomics fields. Certainly, there are many other powerful cancer-oriented bioinformatics and cheminformatics resources and algorithms that are not covered in this review but comprehensively overviewed elsewhere.73–75 For example, UCSC Xena (https://xena.ucsc.edu) is a powerful server to explore cancer genomics and clinical data. The ProbeMiner server (https://probeminer.icr.ac.uk) complements the functionality of Probes and Drugs portal and the Small Molecule Suite tools to identify chemical compounds with pharmacological activity against specific target proteins. Furthermore, the NCI CTD2 server provides a unique collection of the cutting-edge cancer-focused analytical tools: https://ocg.cancer.gov/programs/ctd2/analytical-tools.

There is no doubt that simultaneously with the development of new experimental approaches, the computational methods and resources will continue to play a vital role in target and drug discovery to accelerate the development of new therapeutic strategies for cancer treatment. Meanwhile, the exponentially growing number of individual tools, servers, and databases makes it practically challenging to stay updated with constantly evolving algorithms and the most efficient informatics resources. The NCI-initiated resources, such as the CTD2 Dashboard63 and CellMinerCDB, as well as the cBioPortal or more recently launched OncoPPi Portal, illustrate the major move toward the integration of diverse cancer-focused multi-omics data from different sources into single “hub” servers. We envision, that for the next decade the development of comprehensive data-integrating resources will be the major direction in computational target and chemical probe discovery to facilitate cancer therapeutic development.

Conflicts of interest

There is no conflict of interest to declare.