Abstract 5063: Predicting target driven toxicity for small molecule inhibitors to aid in drug development

Abstract Adverse events and toxicity concerns account for roughly 50% of clinical trial failures and the majority of drug withdrawals. A large part of these toxicity effects can be explained by a drug’s molecular target, known off targets or a drug’s lack of specificity. Therefore, the understanding of a target’s potential toxicity is crucial in the drug development process, potentially informing therapeutic hypotheses and/or chemical design. Here, we introduce a machine learning model to predict a gene’s likelihood to result in patient toxicity when inhibited, independent of chemical compound information. We used information on drug approvals, discontinuations and withdrawals from the FDA to define “toxic” and “safe” targets, to be used as training data. Our model leveraged preclinical, gene-specific features that can be easily collected such as tissue expression, pathway involvement, and loss of function data, and achieves significant predictive performance (ROC = 0.80). We validated our model’s ability to discriminate between successful and unsuccessful drug candidates based on their predicted target toxicity. Furthermore, our model was able to predict differences in toxicity between gene isoforms, which were in agreement with published in vivo evidence, further highlighting its ability to be applied to guide drug development and lead optimization. By applying our model, the prediction of target-specific toxicity can guide drug screenings, prioritize candidates and avoid costly and dangerous drug failures. Citation Format: Manuel Garcia-Quismondo, Mukti Parikh, Neel Madhukar, Coryandar Gilvary. Predicting target driven toxicity for small molecule inhibitors to aid in drug development [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5063.