Abstract LB069: Predictive modeling of drug sensitivity based on the genetic dependency

Functional genetic screens have been frequently used to repurpose existing drugs or to identify new drug targets. However, the associations between hits on screens and candidate drugs have been unclear, and a manual approach to directly infer drug candidates from “druggable” candidates suffers from its laborious, ad hoc investigation and low accuracy. Here, we describe a systematic approach to explore the associations between genetic dependencies and drug sensitivity of cancer cell lines and to identify potential drug repositioning opportunities. We first developed drug sensitivity predictive models trained on genome-wide CRISPR-Cas9 knockout viability profiles and large-scale drug response screens of 4,518 drugs from DepMap. We obtained the highest predictive power from the decision-tree-based models and observed previously reported associations between gene perturbation and drug response. For example, cancer cells that are more sensitive to TP53 or MDM2 perturbation were more sensitive to MDM inhibitors, and EGFR or ERBB2 dependency strongly predicted the sensitivity of cells to EGFR inhibitors. Interestingly, for 3,260 drugs with known targets, only 1.5% of these genes were selected as top features to predict drug sensitivity, indicating that genes that are not the direct targets of drugs should be considered. In addition, we applied our models to the CRISPR-Cas9 knockout screens that target 1,805 “druggable” genes and predicted the sensitivity of cancer cells to known and novel drugs. Furthermore, we were able to validate the candidate drugs using cell viability assays. Finally, a multi-class classification deep learning model, based on the TorchDrug framework, reaches higher recall values than a decision-tree-based method but with much lower precision values. Our study provided the first systematic evaluation between large-scale gene dependency and drug responses. The machine learning predictive model provides an in-silico approach to perform drug screens from functional genetic screens. Citation Format: Bicna Song, Lumen Chao, Xiaolong Cheng, Yuan Gao, Ruocheng Shan, Wei Li. Predictive modeling of drug sensitivity based on the genetic dependency [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 2 (Clinical Trials and Late-Breaking Research); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(8_Suppl):Abstract nr LB069.