Abstract 3688: Target identification for anticancer molecules using a Big Data approach
Neel S. Madhukar, Linda S. Huang, Kaitlyn M. Gayvert, David S. Rickman, Olivier Elemento · Cancer Research · 2015
Abstract Identifying the targets of bioactive compounds has been a major challenge, with efforts being driven by case specific experimentation - a slow and failure-prone process. Various methods have attempted to facilitate discovery, but frequently rely on unavailable cohorts of known binding ligands or complex 3D structures. On the other hand, there has recently been an increase of large-scale genomic, chemical, and structural datasets. By integrating these, molecules can be represented as over 100,00 unique data points, and we hypothesized that combining these various sources into a comprehensive prediction engine could radically improve our ability to predict the drug targets and identify novel anticancer compounds. To test this hypothesis, we developed BANDIT - a Bayesian Approach to find Novel Drug Interaction Targets. BANDIT integrates datasets on drug efficacies, post-treatment transcriptional responses, drug structures, known adverse effects, and bioassay sensitivities, in order to compute an overall Bayesian likelihood and predict drugs that may share a biological target. When applied to a test set of known drugs, BANDIT achieved a predictive power of 89% at identifying drug pairs known to share a target (AUROC = 0.89). Across these datasets we observed that the most predictive were structural similarity and similar survival responses across various cell lines. Moreover the power of the classifier steadily increased from 60% to 89% as the number of included data types was increased- indicating the strength of BANDIT's Big Data approach. We then used BANDIT to determine drugs that could be used for cancer therapy. Specifically, we looked for non-cancer drugs predicted to share a target with known anticancer drugs and for novel targets of known anticancer drugs. We predicted two common cancer drugs, Resveratrol and Genistein, to share a target (Likelihood ratio (LR) = 79.8; Top .05% of predictions). Studies have shown that resveratrol enhances the apoptotic effect of Genistein and our prediction reveals that a possible mechanism for the additive effect could be the dual inhibition of a single target. We also predicted the unreported potential of Vismodegib - used to inhibit the Hedgehog signaling pathway in basal cell carcinoma - to act as a tyrosine kinase inhibitor (LR = 348.1; Top .01% of predictions). Additionally, microtubule-targeting drugs have been important chemotherapy agents, and we predicted Mebendazole and Romidepsin to both inhibit tubulin formation (Likelihood ratios = 178.3 & 210.1 respectively; Top .02% of predictions), thus presenting the possibility for them to be used in an anti-tubulin chemotherapeutic context. Altogether, BANDIT provides a novel, broadly applicable way to identify novel targets for new and established drugs and could greatly expedite pharmaceutical research. Moreover, it could significantly impact therapeutic decisions by determining drugs that could be repositioned to target important cancer regulatory proteins. Citation Format: Neel S. Madhukar, Linda Huang, Kaitlyn Gayvert, David Rickman, Olivier Elemento. Target identification for anticancer molecules using a Big Data approach. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 3688. doi:10.1158/1538-7445.AM2015-3688