Abstract 4910: AI predicts drug response from the genomic features of cells and the kinase activity changes induced by compounds

Abstract

Abstract Human protein kinases are playing an important role in various biological activities including tumorigenesis and cancer progression, and for this reason, many kinase inhibitors have been developed as promising targeted therapies for cancer. To date, there are 72 FDA-approved treatments targeting about two dozen different protein kinases; 3 of them were approved in 2022. However, there are still critical challenges that can be addressed to improve the robustness and selectivity of the kinase inhibitors such as drug resistance, toxicity and compromised efficacy. To gain a deeper understanding of the biological mechanisms underlying the resistance to kinase inhibitors, we hereby design and train an artificial intelligence (AI) model that predicts in-vitro cell response to various kinase inhibition from the integration of the genomic features of cancer cell lines and the biochemical profiles of kinase activities. The inner-working of an optimized model provides important insights on the relevant kinase activities for the cell response. A major challenge for training such a model is that the training data for biochemical kinase activity is very sparse; the kinase panels are not consistent across different sources and each dataset contains many missing values. To address this, we train a secondary AI model to predict all the missing values for the kinase activity in our training data using the combination of the compound chemical structure and the kinase structure. The secondary model is independently validated to ensure the quality of the training data for the drug response prediction model. An investigation of the secondary model’s inner mechanism sheds light on the kinase binding sites and which chemical substructures are important for its inhibition. The model accurately predicts the cell response to different CDK inhibition. A focused analysis of the model characterizes off-target effects of the CDK inhibitors which may be linked to the drug resistance mechanisms, suggesting ways to improve the inhibitors or potential targets for combination therapies. Citation Format: Rens Janssens, Kimberly Rizzolo, Erin Artin, Joel Wagner, Markus Riester, Joshua Korn, Jisoo Park. AI predicts drug response from the genomic features of cells and the kinase activity changes induced by compounds [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4910.