Community assessment to advance computational prediction of cancer drug combinations in a pharmacogenomic screen

Michael P Menden ,Jonathan R Dry ,In Sock Jang ,Mathew J Garnett ,Bence Szalai ,Mehmet Eren Ahsen ,Russ Wolfinger ,Tin Nguyen ,Mike J Mason ,Krishna C Bulusu ,Jaewoo Kang ,Stephen E Fawell ,Gustavo Stolovitzky ,Robert Vogel ,Thea Norman ,Elias Chaibub Neto ,Mikhail Zaslavskiy ,Minji Jeon ,Yuanfang Guan ,Justin Guinney ,Giovanni Y Di Veroli ,Dennis Wang ,Thomas Yu ,Jing Tang ,Zara Ghazoui ,Julio Sáez-Rodríguez

doi:10.1038/s41467-019-09799-2

Abstract

The effectiveness of most cancer targeted therapies is short-lived. Tumors often develop resistance that might be overcome with drug combinations. However, the number of possible combinations is vast, necessitating data-driven approaches to find optimal patient-specific treatments. Here we report AstraZeneca’s large drug combination dataset, consisting of 11,576 experiments from 910 combinations across 85 molecularly characterized cancer cell lines, and results of a DREAM Challenge to evaluate computational strategies for predicting synergistic drug pairs and biomarkers. 160 teams participated to provide a comprehensive methodological development and benchmarking. Winning methods incorporate prior knowledge of drug-target interactions. Synergy is predicted with an accuracy matching biological replicates for >60% of combinations. However, 20% of drug combinations are poorly predicted by all methods. Genomic rationale for synergy predictions are identified, including ADAM17 inhibitor antagonism when combined with PIK3CB/D inhibition contrasting to synergy when combined with other PI3K-pathway inhibitors in PIK3CA mutant cells.

Highlights

The effectiveness of most cancer targeted therapies is short-lived
Sparsity in the cell line drug combination data matrix (Fig. 1g) resulted as several drug combinations were selectively profiled in clinically relevant cancer cell lines, e.g., ESR1 inhibitors were predominantly combined with other drugs in estrogen receptor-positive breast cancer cell lines since these agents are standard of care within this cancer subtype
In the 21 O’Neil et al.[4] cell lines not used within AZ-Dialog for Reverse Engineering Assessments and Methods (DREAM) training, 11 of the prioritized biomarker-drug combinations were present, of which 8 (72%) showed reproducible directionality (Fig. 7d). The objective of this AstraZeneca-DREAM Challenge was to drive the development of innovative computational approaches to predict novel drug combinations and to comprehensively benchmark these approaches

Summary

Introduction

The effectiveness of most cancer targeted therapies is short-lived. Tumors often develop resistance that might be overcome with drug combinations. We report AstraZeneca’s large drug combination dataset, consisting of 11,576 experiments from 910 combinations across 85 molecularly characterized cancer cell lines, and results of a DREAM Challenge to evaluate computational strategies for predicting synergistic drug pairs and biomarkers. While empirical experiments are important for observing potential synergistic properties across drug pairs, the possible number of combinations grows exponentially with the number of drugs under consideration This is further complicated by the influence of disease and cellular contexts, rendering it impractical to cover all possibilities with undirected experimental screens[4]. A number of approaches have been developed to model drug combination synergy using chemical, biological, and molecular data from cancer cell lines[6,7] but with limited translatability to the clinic. DREAM Challenges (dreamchallenges.org[www.dreamchallenges.org]) are collaborative competitions that pose important biomedical questions to the scientific community, and evaluate participants’ predictions in a statistically rigorous and unbiased way, emphasizing model reproducibility, and methodological transparency[9]

Methods

Results

Conclusion