Development and external validation of an interpretable QSAR model for quinolinylchalcone derivatives against MCF-7 breast cancer cells

In this work, an interpretable QSAR model was developed to predict the antiproliferative activity of quinolinylchalcone derivatives against the Michigan cancer foundation-7 (MCF-7) breast cancer cell line. In order to improve the chemical consistency of the model, the dataset was limited to 42 compounds derived from related series such as quinoline–chalcone, chalcone-fused quinoline, and quinoline/chalcone/1,2,4-triazole hybrids. The biological activity was expressed in terms of pIC50, then correlated with 2D molecular descriptors. The final linear model is based on three descriptors: FractionCSP3, NumHAcceptors, and NumAromaticHeterocycles, with acceptable internal robustness: R 2 = 0.658, Q 2 LOO = 0.584, and Q 2 CV5 = 0.575. The external validation, conducted on six independent quinolinylchalcones, yielded R 2 ext = 0.496, RMSEext = 0.328, and MAEext = 0.301. The applicability domain analysis showed that four out of six external compounds fall within the model’s domain. These results indicate that refocusing on a chemically homogeneous sub-family improves the chemical coherence and interpretability of the model, while its external predictive performance remains moderate. The proposed model thus constitutes a useful exploratory tool to guide the design of new quinolinylchalcone derivatives with anti-MCF-7 activity. In addition, docking analysis provided a complementary structure-based interpretation, showing favorable accommodation of the selected compounds within the EGFR ATP-binding pocket. The selected compounds were predicted to form stabilizing hydrogen-bond, π-related, and hydrophobic interactions within the EGFR binding pocket.