Abstract 4278: Transcriptome-driven combination design: A computational approach

Abstract

Abstract While rational drug combination design seeks to improve patient outcomes by leveraging biological synergy, outcomes from combination trials often fail. Only 12.4% of phase 1 trials eventually achieve regulatory approval. In vitro evidence has provided the rationale for the drug combinations in 36.6% of phase 1 combination trials, but none of those trials progressed past phase 3 (from a study by Paller CJ et al., 2019). A review of DrugCombDB, a publicly available database of experimental synergy results, supports the idea that such outcomes are mirrored in preclinical data. Analysis of the similarity of synergy scores within indication-specific subsets of cell lines from data in a 2022 paper by Douglass, et al. show that in both breast and lung cancer, combinations predicted to be synergistic in at least one cell line from that indication failed to be synergistic in the majority of indication cell lines more than half of the time. SHEPHERD’s Gene Cluster Voting Algorithm (GCVA) was adapted to improve upon current drug combination design and anticipate transcriptome changes as the result of drug treatment. GCVA relates drug sensitivity to the transcriptome by calculating a list of biomarkers that contribute to sensitivity and resistance for specific drugs, enabling the prediction of drug sensitivity on patient transcriptomic data captured via bulk RNAseq. Via this method, 15,137 publically available synergy data points in the Douglass paper were compared to predictions by our computational approach. For all synergy metrics contained within the dataset (ZIP, Bliss, Loewe, and HSA), the synergistic predictions made by application of GCVA have significantly higher synergy scores compared to the antagonistic predictions made by GCVA (p < .001). These test results support the potential for the application of GCVA to intelligent combination design with the ultimate goal of deployment on a patient-by-patient basis.In addition, analysis of GCVA’s drug efficacy predictions for the cell line DU145 showed that treatment with 9 kinase inhibitors consistently potentiated the predicted efficacy of between 5 and 11 additional drugs. Dasatinib, temsirolimus, and cabozantinib were newly predicted effective after treatment with each and every kinase inhibitor. This data suggests an avenue for the design of intelligent drug combinations based on anticipated transcriptomic changes that may improve on existing synergy prediction methods. Citation Format: Christina Gavazzi, Mikhail G. Grushko, Jeremy M. Goldstein, Mahta Samizadeh, Zakary Y. ElSeht, Katherine Arline. Transcriptome-driven combination design: A computational approach. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4278.