Abstract B076: A DREAM-world of approaches for predicting combination synergies in cancer

Abstract Selecting the right drug combination for the right patient is a complex problem. Some success has been seen from preclinical screening and computational approaches combining molecular, drug and tumor properties to predict combination benefit within the tumor cell. However the context specificity and heterogeneity of drug resistance is often overlooked, leading to inappropriate assumptions of statistical power and resulting in poor translatability. Knowledge-aware approaches will be needed to augment pre-clinical statistical discovery but a huge variety of techniques exist, each offering different strengths. To test this world of methods, AstraZeneca partnered with DREAM, Sage Bionetworks and the Sanger Institute. We released over 11,000 test results from drug combinations in cancer cell lines for an incentivized crowd-sourcing challenge to develop computational models predicting synergistic drug combinations and biomarkers determining response. This became the highest participated of any DREAM challenge, with >700 participants from diverse geographies and knowledge/skill specialties collectively contributing >50,000 hours to the problem. Over 70 models were submitted, testing most machine learning algorithms alongside knowledge-driven approaches. The best models out-performed the state of the art, consistently predicting at an equivalent level to biological replication. Network techniques to select molecular markers most relevant to drug target biology improved prediction the most. Amongst results that could be validated in independent data, ERBB2/EGFR/EP300 variants were predictive for combinations inhibiting AKT+EGFR/ERBB, KRAS for BRAF+MAP2K, AR for AKT/SGK+MAP2K, and ESR1 for ESR1+AKT/PI3K. Alongside revealing interesting biology in our data and providing predictive models, the challenge has given clear direction to our future investment in computational methods for combination pharmacogenomics. Citation Format: Jonathan R. Dry, Michael Menden, Julio Saez Rodriguez, Justin Guinney. A DREAM-world of approaches for predicting combination synergies in cancer [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference: Molecular Targets and Cancer Therapeutics; 2017 Oct 26-30; Philadelphia, PA. Philadelphia (PA): AACR; Mol Cancer Ther 2018;17(1 Suppl):Abstract nr B076.