H-RACS

H-RACS predicts synergistic drug combinations for cancer treatment by ranking candidate pairs using predictive models trained on large-scale combinational data.

Key Features:

• Data-driven predictions: Predictive models were trained on a dataset comprising 33,574 combinational scenarios.
• Minimal input requirements: Predictions rely on chemical structures and target information of candidate drugs and do not require drug–cell treatment results.
• Cancer context coverage: Models incorporate a built-in context of 928 cell lines across 24 cancer types.
• Predictive performance: Achieved an AUC of 0.89 on independent combinational scenarios and 67% precision within the top 5% ranking on the DREAM dataset.
• Extendibility: Maintains high accuracy on new data with AUCs of 0.84 for new drug combinations and 0.81 for new cell lines.

Scientific Applications:

• Pre-screening of drug pairs: Prioritizes candidate drug combinations for experimental testing without prior treatment assays.
• Oncology research prioritization: Ranks combinations to guide follow-up validation studies in cancer models.
• Extension to novel agents and cancers: Supports evaluation of new drugs and cancer types using the built-in cell line context.
• Support for targeted and personalized strategies: Informs selection of combination strategies relevant to specific cancer contexts.

Methodology:

Predictive models were trained on 33,574 combinational scenarios using chemical structure and target information as inputs; models incorporate data from 928 cell lines across 24 cancer types and were validated on independent combinational scenarios and the DREAM dataset reporting AUC and precision metrics.