Bioactivity-driven prediction of antibacterial synergy using machine learning models.

Machine learning model HALO predicts antibacterial drug synergy with up to 0.90 ROC-AUC across independent datasets, demonstrating robust generalization.

• Why it matters: Accurate prediction of antibacterial synergy is crucial for developing effective combination therapies, but existing models often overestimate performance due to flawed evaluation methods.
• What they did: A curated dataset of 3,160 drug-strain interactions was used to develop HALO, which encodes drug pairs with multi-level Chemical Checker bioactivity features and employs strict nested cross-validation.
• The result: HALO achieved consistent, high-performance predictions under rigorous testing, highlighting the value of bioactivity signatures and setting realistic expectations for antibacterial synergy modeling.