Bioactivity-Driven Prediction of Antibacterial Synergy Using Machine Learning Models

Predicting antibacterial drug synergy remains difficult due to strain variability and the limited scale of experimentally tested combinations. Existing machine-learning approaches often rely on permissive cross-validation schemes that allow drug pairs to appear across folds, inflating performance. A rigorous evaluation framework and scalable feature representation are needed for robust generalization. We assembled a curated dataset of 3,160 drug–pair–strain interactions covering 97 compounds and 10 bacterial strains. We then developed HALO (Held-out Antibiotic interaction Learning from latent bioactivity Observations), a synergy-prediction framework in which each drug pair is encoded using multi-level Chemical Checker (CC) similarity features spanning chemical, target, network, cellular, and clinical bioactivity domains. Under strictly nested, pair-heldout cross-validation (CV1), HALO achieved stable generalization to unseen combinations (accuracy ≈ 0.75; ROC–AUC = 0.82). Performance depended strongly on evaluation stringency: models performed well under random splits but degraded when required to generalize to unseen drug pairs and strain contexts. Despite these constraints, HALO generalized to an independent set of Loewe-α measurements, achieving ROC–AUC = 0.85 for distinguishing synergy from antagonism. These results demonstrate that multi-level bioactivity signatures provide a scalable, interpretable basis for predicting antibacterial synergy and reveal the performance limits of current models under rigorous evaluation. Code, data-processing scripts, and trained models will be available at GitHub repo. [email protected] Supplementary figures and additional evaluation details are available online.