CLC-Pred Synergy: Web Application for Predicting Pairwise Drug Combinations with Synergistic Activity Against NCI60 Cancer Cell Lines

Pharmacological intervention is central to cancer therapy, but overcoming drug resistance is limited by severe adverse effects. Rational drug combinations can improve efficacy and reduce toxicity, yet identifying effective pairs is challenging due to the vast combinatorial space. Structure-activity relationship (SAR) modeling offers an in silico approach to predict drug synergy. To create SAR models, training sets were built from the NCI-ALMANAC database, containing cytotoxicity profiles of ~5000 drug pairs per NCI-60 cell line. SAR models were developed using the PASS DDI (Prediction of Activity Spectra for Substances with the Drug-Drug Interaction) platform, which encodes chemical structures into PoSMNA (Pairs of Substances Multilevel Neighborhoods of Atoms) descriptors and applies a Bayesian-like algorithm to uncover structure–activity patterns. Model performance was assessed using leave-one-out compounds out cross-validation (LOO CO CV), with predictive accuracy quantified by the area under the ROC curve (AUC). A total of 104 SAR models demonstrated robust predictive performance, with AUC values exceeding 0.7 (mean AUC = 0.75), enabling the prediction of synergistic effects for compound combinations across 45 cancer cell lines. These models have been implemented in the CLC-Pred Synergy web application, providing a practical tool for in silico screening of anticancer drug combinations.