Predicting and Optimizing Synergistic Drug Combinations for Breast Cancer Treatment Using Machine Learning.

Objectives: The study aims to identify highly synergistic drug combinations for breast cancer treatment using machine learning models. The primary objective is to predict drug synergy scores accurately and rank combinations with the highest potential for therapeutic efficacy.

Methods: Machine learning models, including XGBoost, Random Forest (RF), and CatBoost (CB), were employed to analyze breast cancer drug combination data. Four synergy metrics-ZIP, Bliss, Loewe, and HSA-were used to quantify drug interaction effects. The models were trained to predict these synergy scores, and their performance was evaluated using normalized root mean squared error (NRMSE) and Pearson correlation coefficient. Predicted top-ranking drug combinations were further validated by comparing observed versus expected dose-response curves and calculating the area under the curve (AUC) for synergy assessment.

Results: XGBoost (XGB_5235) outperformed other models, achieving an NRMSE of 0.074 and a Pearson correlation of 0.90 for the Bliss synergy model. Based on average synergy scores, the top 20 drug combinations were identified, with Ixabepilone+Cladribine, SN 38 Lactone+Pazopanib, and Decitabine+Tretinoin emerging as the most promising. These combinations showed high synergy and were supported by biological insights into their mechanisms of action.

Conclusions: The study demonstrates the effectiveness of machine learning in predicting synergistic drug combinations for breast cancer. By accelerating the screening process and reducing experimental burden, the approach offers a promising tool for guiding future in vitro and in vivo validation of combination therapies.