Abstract Synthetic lethality exploits the phenomenon that a mutation in a cancer gene is often associated with new vulnerability which can be uniquely targeted therapeutically, leading to a significant increase in favorable outcome. DNA damage and survival pathways are among the most commonly mutated networks in human cancers. Recent data suggest that synthetic lethal interactions between a tumor defect and a DNA repair pathway can be used to preferentially kill tumor cells. We recently published a method, DiscoverSL, using multi-omic cancer data, that can predict synthetic lethal interactions of potential clinical relevance. Here, we apply the generality of our models in a comprehensive web tool called Synthetic Lethality Bio Discovery Portal (SL-BioDP) and extend the cancer types to 18 cancer genome atlas cohorts. SL-BioDP enables a data-driven computational approach to predict synthetic lethal interactions from hallmark cancer pathways by mining cancer’s genomic and chemical interactions. Our tool provides queries and visualizations for exploring potentially targetable synthetic lethal interactions, shows Kaplan–Meier plots of clinical relevance, and provides in silico validation using short hairpin RNA (shRNA) and drug efficacy data. Our method would thus shed light on mechanisms of synthetic lethal interactions and lead to the discovery of novel anticancer drugs. Keywords: cancer, synthetic lethality, web application, bioinformatics, DNA repair pathway 1. Introduction Drug treatment of cancer depends on the notion that mutations that give rise to the development of cancer also bring about a weakness that can be exploited therapeutically. Large-scale cancer genome sequencing efforts have catalogued mutations in various cancer types that can be explored as tumor-specific vulnerabilities [[32]1]. These genetic alterations consist of gain-of-function mutations in which genes are amplified, translocated, or mutated and loss-of-function mutations in which gene function is compromised by missense mutations or deletions. The former group of mutations have been the subject of intense focus by the pharmaceutical industry for the development of targeted cancer drugs. These efforts have resulted in several cancer drugs that target activated driver oncogenes, such as HER2, BCR-ABL, EGFR, and BRAF [[33]2]. These drugs target signaling proteins that are aberrantly activated as a direct consequence of an oncogenic mutation, and hence their inhibition is detrimental to the cancers. This dependence on oncogenic driver pathways is commonly referred to as oncogene addiction [[34]3]. From a drug discovery perspective, the loss-of-function mutations are much harder to tackle, and the same is true for several activated oncogenes that have proven to be undruggable, such as the MYC transcription factor and the RAS proteins. Therefore, alternative strategies are needed to target the vulnerabilities induced by these classes of cancer-causing genes. One promising way to tackle this challenge is based on the concept of synthetic lethality (SL). SL describes the relationship between two genes whereby inactivation of either gene alone results in a viable phenotype, while their combined inactivation is lethal. SL has long been considered a foundation for the development of selective anticancer therapies [[35]4,[36]5,[37]6,[38]7,[39]8], which aim to inhibit the SL partner of a gene that is inactivated de novo in the cancer cells. Beyond guiding the development of novel selective cancer therapies, it has been noted that the network of SL interactions can give a bird’s eye view of the genomic state of a given tumor that can be used to find tumor-specific vulnerabilities and develop effective synergistic drug combination therapies in a precision-based manner [[40]9,[41]10]. Given the importance of SL, considerable work has been devoted to finding such interactions in cancer—both experimentally [[42]4,[43]5,[44]6,[45]7,[46]8] and computationally [[47]11,[48]12]. Nevertheless, so far, the utility of SL in the clinic has been limited, and many of the SLs found in current screens manifest a poor predictive signal in actual patients’ data. A recent publication tried to bridge this gap by finding the clinically relevant SL interactions from cell-line based SL screens [[49]13]. But again, cell-line based SL screens have been done for only a limited set of cancer genes and cell lines. So, there is a need for a comprehensive resource that can be queried for alternative drug targets for important cancer genes based on the concept of synthetic lethality, which shows potential to be clinically relevant. A previously published database, namely, SynLethDB [[50]14] has a collection of synthetic lethal partners from multiple sources (text mining, synthetic lethality screens, and computational prediction), but this database lacks the parameters to assess the clinical relevance of these SL interactions. Here, we present an integrative web portal, Synthetic Lethality BioDiscovery Portal (SL-BioDP), that enables multilevel querying and visualization of synthetic lethality-based potential drug targets for genes which are frequently mutated in cancers. To explore the possibilities of precision therapies for specific gene alterations using the concept of synthetic lethality, we inferred SL interactions of known cancer-driver genes or hallmark cancer pathways. A published statistical approach [[51]15] was used to identify potential SL interaction from 18 different cancers from the TCGA cohort (The Cancer Genome Atlas) in a genome-scale manner, and assess their clinical relevance. Combined with in silico validation from large-scale in vitro drug response and shRNA screens, our resource gives a venue to query and visualize drug targets, and allows hypothesis generation for personalized medicine. 2. Results 2.1. Data Summary SL-BioDP predicts synthetic lethal partners of reported cancer driver genes and genes from hallmark cancer pathways in 18 cancer types. Cancers are mainly caused by mutations/alterations of specific genes or pathways that contribute to the initiation and progression of the disease. In our study, we aimed to find potentially targetable and clinically relevant synthetic lethal partners for reported cancer driver genes [[52]16] and genes belonging to 10 hallmark cancer pathways [[53]17] in 18 types of cancers from TCGA (as shown in [54]Figure 1a,b). For the synthetic lethality prediction model in each cancer type, we only included genes which are mutated in at least five tumor samples in TCGA. [55]Figure 1a,b shows the number of mutated driver genes and the number of genes from 10 cancer pathways included in each of the 18 cancer types, which totals 623 unique primary genes. The synthetic lethal partners for the primary genes in the cancer types are predicted using DiscoverSL [[56]15]. For cancer-specific synthetic lethal interactions of each of the 623 primary genes reported in SL-BioDP, the DiscoverSL algorithm was run on the whole genome-scale somatic mutation, expression, copy-number alteration, and clinical data of 7654 tumor samples from 18 TCGA cancer types. Published reports of interactions from SL-screening data were collected from two sources [[57]13,[58]14] and overlapped with the predictions from DiscoverSL. [59]Table 1 shows the number of SL interactions for each cancer type either supported by Achilles short hairpin RNA (shRNA) screenings or from literature. Outside the database, we tested an additional in silico validation from CRISPR (clustered regularly interspaced short palindromic repeats) knockout data from project Achilles (DepMap Public 19Q3) on 12 selected cancer driver genes frequently mutated in cancers: BRCA1, BRCA2, TP53, PTEN, ATM, ATR, KRAS, HRAS, BRAF, EGFR, MET, and PIK3CA. Many of the predicted SL interactions of these 12 genes across 18 cancers were also present in the genetic dependency CRISPR screen (shown in [60]Supplementary Table S1). A pathway enrichment analysis for SL partners for a primary gene in a cancer type can be viewed. Finally, drugs targeting the synthetic lethal partners can be explored for a chosen list of genes. For SL-based drug target searching, gene–drug interactions from the two databases, DrugBank [[61]18] and DGIdb [[62]19], were incorporated which total 20,899 and 28,104 gene–drug interactions, respectively. Figure 1. [63]Figure 1 [64]Open in a new tab General schema of SL-BioDP: (a) The bar-chart shows the number of cancer driver genes in 18 cancer types included as primary genes in SL-BioDP; (b) the heatmap shows the number of genes from each of the 10 hallmark cancer pathways included as primary genes in SL-BioDP for each of the 18 cancer types. The color coding is based on the frequency of the number of genes from a pathway in each cancer (red: more number of genes, green: less number of genes). Table 1. The number of SL pairs in each cancer type that passes either of the two validation steps: Conditional sensitivity from RNAi (RNA interference) screening data in cancer cell lines or reported in literature. Cancer Validated from RNAi Screen in Cancer Cell Lines (p-Value < 0.05) Validated from Literature Bladder urothelial carcinoma: BLCA 8257 337 Breast invasive carcinoma: BRCA 37,224 1095 Cervical squamous cell carcinoma: CESC 8198 846 Colorectal adenocarcinoma: COADREAD 2681 972 Glioblastoma multiforme: GBM 1918 187 Head and neck squamous cell carcinoma: HNSC 37,852 1917 Kidney renal cell carcinoma: KIRC 7566 584 Acute myeloid leukemia: LAML 291 4 Lower grade glioma: LGG 1041 55 Liver hepatocellular carcinoma: LIHC 16,731 1430 Lung adenocarcinoma: LUAD 15,767 1148 Lung squamous cell carcinoma: LUSC 8012 568 Ovarian serous cystadenocarcinoma: OV 619 111 Pancreatic adenocarcinoma: PAAD 2240 458 Prostate adenocarcinoma: PRAD 869 185 Skin cutaneous melanoma: SKCM 52,985 2171 Thyroid carcinoma: THCA 132 74 Uterine corpus endometrial carcinoma: UCEC 76,090 2915 [65]Open in a new tab Summary of the data included in SL-BioDP is shown in [66]Figure 2. The circos plot in [67]Figure 2a shows an example of selected primary genes common in multiple cancer types. Each sector in the circos plot corresponds to a cancer type and the x-axis labels show the primary genes included in the cancer type. For each primary gene in a cancer type, information on the number of predicted synthetic lethal genes, number of mutated samples, number of drugs targeting synthetic lethal genes, and number of enriched synthetic lethal pathways are shown in tracks 1, 2, 3, and 4, respectively (see [68]Supplementary Tables S2 and S3). Depending on the gene expression, copy number, and mutation profiles of the genes in each cancer types, synthetic lethality of gene pairs can be cancer-specific. However, we found a common synthetic lethal signature in multiple cancers for some very commonly mutated cancer genes: ATM, NF1, PIK3CA, PRKDC, RB1, and TP53. [69]Figure 2b shows a map of common synthetic lethal pairs (common in at least 10 cancer types) in cancers, grouped by the primary genes. To support the idea of precision therapeutic approaches for treating cancer patients having specific genetic alterations, we demonstrate the utility of SL-BioDP for suggesting SL-based drugs depending on the presence of mutations in specific cancer genes. From the SL partners of frequently mutated cancer driver genes, we inferred the common drugs (used for cancer treatment in the National Cancer Institute (NCI)) targeting these SL partners from DrugBank and DGIdb. These drugs can be potential treatment choices for treating patients carrying mutations in their corresponding primary genes in multiple types of cancers ([70]Figure 2c). Figure 2. [71]Figure 2 [72]Open in a new tab Summarization of cancer-specific synthetic lethality data presented in SL-BioDP: (a) Circos plots displaying the summary of synthetic lethal interactions of 63 cancer genes that are common in at least five cancer types. The outermost layer (layer 1) shows the number of predicted synthetic lethals per gene. The next layer (layer 2) shows histograms of the number of mutations per gene. The next to last layer (layer 3) shows the number of drugs identified per gene, through their synthetic lethal partners. Finally, the innermost layer (layer 4) displays histograms of the number of enriched synthetic lethal pathways per gene; (b) the table shows all common synthetic lethal pairs with high prediction scores (>0.7) in at least 10 cancer types. For each synthetic lethal pair (columns) and each cancer type (row), the color coding shows whether that SL pair has a high prediction score (>0.7) in that cancer type (color: blue) or not (color: white). The SL pairs in columns are grouped by their primary genes (ATM, NF1, PIK3CA, PRKDC, RB1, and TP53); (c) similar to the table in (b), this table shows the common SL-based drugs targeting the common SL partners of the primary genes (ATM, NF1, PIK3CA, PRKDC, PTEN, and TP53) in at least 10 cancer types. 2.2. Searchability and Browsing Users can search SL-BioDP from any of the three modes: (1) “GENES” tab: using a primary gene of interest (the model currently includes 623 cancer genes commonly mutated in cancers), (2) “CANCER” tab: tumor tissue of origin (currently includes 18 cancer histology types), and (3) “DRUG” tab: drugs of interest (drug targets are compiled from the two databases, DrugBank and DGIdb). Additionally, potential synergistic drug combinations for different cancers can be browsed from the “INFERRED DRUG SYNERGY” tab. An example of search and browsing functionalities in SL-BioDP is illustrated in [73]Figure 3. Figure 3. Figure 3 [74]Open in a new tab The utility of SL-BioDP is shown with a case study on BRAF in cancer lung adenocarcinoma (LUAD). Using “GENE” search in SL-BioDP, the mutation and expression alteration profile of BRAF across 18 tumor histology types is shown. BRAF is commonly mutated in LUAD (4.26%). Otherwise, using the “CANCER” search in SL-BioDP and choosing LUAD as the histology, genes having the most significant alteration in expression (comparing tumor vs. normal samples from TCGA) are displayed. BRAF shows more than 6-fold upregulation in tumor samples (p-value < 0.001). When searched for SL interactions, BRCA1 is a predicted SL partner of BRAF in LUAD (synthetic lethal score = 0.87 and survival p-value = 0.0049). Alternatively, using “DRUG” search in SL-BioDP, searching for the targets of the drug “temozolomide” yields 10 target genes. By selecting the BRCA1 gene and hitting submit, we search for the alteration profiles of BRCA1 across 18 cancers. From the resulting page we can select the histology type (we choose LUAD here) to search for the primary genes whose mutation makes BRCA1 synthetic lethal in the selected cancer type. Restricting the search for reported SL interactions only, we get primary (mutated) genes including BRAF. Survival analysis (Kaplan-Meier plot) shows better disease-free survival in patients with underexpression of BRCA1 compared to overexpression, in LUAD samples with BRAF mutations. Pathway enrichment for the SL interactors of the primary gene (BRAF) can be seen. The drugs targeting either of the SL gene pair is searchable from DGIdb or DrugBank by selection. (1) Search using “GENE” tab: search can be initiated by using either the official gene symbol or Entrez gene id of the primary gene. SL-BioDP presents cancer-specific synthetic lethality information in multiple levels. First, the mutation and expression profiles of the primary genes in each cancer type are presented. Then, the predicted synthetic lethal partners of the chosen primary genes in selected cancer types are presented. The predicted synthetic lethal pairs in SL-BioDP are ranked by the synthetic lethal score calculated by the DiscoverSL model. Additional parameters for assessment of synthetic lethality for each predicted SL pair are shown as: a p-value for expression correlation, mutual exclusivity of mutations, conditional gene essentiality from RNAi screens, and a change in the SL interactor amplification profile in the presence of a primary gene mutation. For validation purposes, to show conditional gene essentiality, we provided an in silico validation approach using shRNA screening data from cancer cell lines [[75]15]. Assessment of the clinical relevance of the predicted SL interactions are shown as Kaplan–Meier survival analyses for disease-free survival and for downregulation vs. upregulation of the predicted synthetic lethal interactor in clinical samples carrying the mutation in the primary gene. In the KM plots, the statistically significant cases (p < 0.05) where downregulation of the synthetic lethal interactor gene shows better disease-free survival in the presence of the mutation in the primary gene are ideal cases of clinically relevant synthetic lethality. Together with the amplification status of the SL interactor gene, we can hypothesize if a favorable clinical outcome can be seen by selective targeting of the interactor gene (Gene2). In addition, we provide reported SL interactions and supporting references of our findings whenever it is