A shape-constrained regression and wild bootstrap framework for reproducible drug synergy testing.

Area of Science:

• Pharmacology
• Biostatistics
• Computational Biology

Background:

• High-throughput drug combination screening is crucial for identifying effective cancer therapies.
• Current synergy scoring methods lack statistical rigor and can fail with complex dose-response data.
• Accurate identification of synergistic drug pairs is essential for drug discovery.

Purpose of the Study:

• To introduce Synergy via Isotonic Regression (SIR), a novel nonparametric framework for analyzing drug synergy.
• To provide a statistically robust method for identifying synergistic drug combinations in high-throughput screens.
• To improve the reliability and accuracy of synergy scoring in drug discovery.

Main Methods:

• Developed a nonparametric framework using 2D isotonic regression to model drug interactions.
• Implemented a degrees-of-freedom-corrected wild bootstrap for calibrated p-value generation.
• Applied SIR to the DrugCombDB database for validation and comparison with existing methods.

Main Results:

• SIR demonstrated superior replicate concordance (median correlation 0.91) compared to baseline methods (0.53-0.74).
• SIR successfully avoided the failure rates associated with Loewe (20.9%) and ZIP (3.6%) synergy scores.
• The SIR model accurately predicted missing experimental data points (median holdout RMSE 0.040).

Conclusions:

• SIR offers a statistically principled approach to drug synergy analysis, replacing heuristic scores with calibrated effect sizes and p-values.
• The framework enhances hit calling accuracy and enables effective error-rate control in large-scale drug screening.
• SIR represents a significant advancement for reliable drug combination discovery.