Abstract 3666: Proteomic-based drug sensitivity prediction algorithms: generation, optimization, and assessment

Abstract Personalized medicine uses genomic information to inform therapeutic decisions. In the past decade, much effort has been made to develop machine learning models to link drug response with large “-omics” expression patterns in pursuit of personalizing medicine. However, until recently, large-scale proteomic datasets have not been widely available and therefore have not often been incorporated into these machine learning methods. Proteomics has the advantage of directly assaying the functional molecular machinery of cells, leading to better understanding of pathway dynamics and quantifying the targets of small molecule inhibitors. Thus, we hypothesize that proteomic data in drug response modeling could offer distinct advantages and better performing drug response models. In this study, we sought to determine if drug response models built on proteomics data can outperform other transcriptomic or multi-omic based models. To this end, we developed and tested a novel framework to predict drug response using only proteomic data. We leveraged several recent large-scale cell line proteomics datasets from DepMap (CTRP) and Sanger (GDSC) to build machine learning models trained on baseline protein expression and drug sensitivity data. We used robust testing to determine optimal parameters, drug sensitivity processing, while comparing performance across multiple machine-learning methods. Our optimized framework demonstrated strong predictive power across multiple drug classes and cancer types. Moreover, our approach successfully predicted clinical outcomes in independent, publicly available datasets, illustrating its translational potential. It was worth noting that across most therapies, our models performed similarly to gene expression-based approaches in identifying responders to both cytotoxic and targeted therapies. The work to improve and optimize these models are on-going with the intent that these tools will be made available to all proteomic researchers looking to directly develop drug response hypotheses from their proteomic data. Citation Format: Robert F. Gruener. Proteomic-based drug sensitivity prediction algorithms: generation, optimization, and assessment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3666.