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

Area of Science:

• Oncology
• Computational Biology
• Pharmacology

Background:

• Cancer targeted therapies often exhibit limited efficacy due to acquired drug resistance.
• Drug combinations offer a strategy to overcome resistance, but the vast number of possibilities requires data-driven approaches.
• Developing patient-specific combination treatments is crucial for improving therapeutic outcomes.

Purpose of the Study:

• To establish a comprehensive dataset of cancer drug combinations and evaluate computational methods for predicting synergy.
• To identify biomarkers associated with synergistic drug responses.
• To benchmark the performance of various computational strategies in predicting drug synergy.

Main Methods:

• A large-scale drug combination dataset was generated, comprising 11,576 experiments across 910 combinations and 85 molecularly characterized cancer cell lines.
• A DREAM Challenge was conducted to assess computational strategies for predicting synergistic drug pairs and biomarkers.
• Winning computational methods integrated prior knowledge of drug-target interactions.

Main Results:

• Computational methods achieved prediction accuracy comparable to biological replicates for over 60% of drug combinations.
• A significant portion (20%) of drug combinations remained poorly predicted across all evaluated methods.
• Genomic rationales for synergy predictions were identified, including specific interactions involving ADAM17 and PI3K pathway inhibitors in PIK3CA mutant cells.

Conclusions:

• Data-driven approaches, particularly those incorporating drug-target interactions, can effectively predict synergistic cancer drug combinations.
• Challenges remain in predicting synergy for certain drug combinations, highlighting areas for future research.
• Understanding the genomic basis of drug synergy is essential for optimizing combination therapies and patient stratification.