Abstract 3896: A big data approach to predicting context specific synergistic drug combinations

Abstract

Abstract Combination therapies for various cancers have been shown to increase efficacy, lower toxicity and escape resistance. However, systematically interrogating all possible synergistic therapies is experimentally unfeasible due to the sheer volume of possible combinations. Computational approaches have proven to be an invaluable tool within pharmacogenomics and have helped with prioritizing the development of perspective therapeutics as well as matching the right drugs with the right patients. Here we apply a novel big data approach in the evaluation and ultimately the prediction of drug synergy by using the recently released NCI-ALMANAC, the largest publically available synergistic drug efficacy dataset to date. First, to better understand drug combinations, we distinguished between those that were synergistic and adverse and evaluated various drug similarity metrics for all pairs. We found that certain features showed significant differences between adverse and synergistic drug combinations, such as post-treatment transcriptional effects similarity (D = 0.17, p < 0.001) and chemical structure similarity (D = 0.25, p < 0.001). By exploiting these significant similarities and dissimilarities and incorporating cell line specific data we developed a machine learning model to predict context specific drug synergy and achieved significant predictive performance (AUC = 0.823). We find that our model can be used to both identify novel synergistic drug pairs, as well as find novel indications for known drug combinations by identifying new sensitive cell lines. Moreover, the interpretability of our model allows for the interrogation of features for a deeper understanding of why certain combinations are predicted synergistic. In addition to identifying the cancer types and subtypes a combination therapy would be most synergistic within, we set out to identify the molecular indication for highly synergistic pairs. Specifically, we systematically identified candidate predictive biomarkers which could be used to stratify patient cohorts. Overall, our model and methodology can expedite the development and expansion of combination therapeutics, which can help battle acquired resistance and increase therapeutic efficacy. The thorough understanding of specific combination efficacy dependencies allows for a true precision medicine application of these therapeutics. Citation Format: Coryandar M. Gilvary, Neel Madhukar, Olivier Elemento. A big data approach to predicting context specific synergistic drug combinations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 3896.