Integrative QSAR and Feature Stability Analysis for Flavonoid-Based Drug Design Targeting A375 Melanoma

Abstract

This study presents a comprehensive QSAR (Quantitative Structure-Activity Relationship) modeling approach for identifying promising flavonoid derivatives that may exhibit anticancer properties against A375 melanoma cells. This is achieved through the application of LASSO (Least Absolute Shrinkage and Selection Operator) regression. Using LASSO regression, we developed a predictive model for biological activity (IC50) based on key molecular descriptors. The model demonstrated strong performance on training data (R² = 0.7445, RMSE = 0.4903) and moderate generalizability in Leave-One-Out Cross-Validation (Q² = 0.5558, RMSE = 0.6465), with no significant overfitting (R² − Q² = 0.1887 < 0.3). Feature stability analysis identified five critical descriptors: LogP (lipophilicity, +0.5340), TPSA (polar surface area, −0.2058), and RotatableBonds (flexibility, +0.1583), alongside NumHDonors and AromaticRings, which collectively explain flavonoid-melanoma interactions. The outcomes are inline with established QSAR patterns, where lipophilicity and polarity influence anticancer activity, where lipophilicity and polarity modulate anticancer potency. Notably, low-frequency descriptors (e.g., MolWeight) were excluded, streamlining future designs. Graphical validation confirmed robust training predictions (R² = 0.7445) and acceptable LOO-CV scatter (Q² = 0.5558), though outliers suggest opportunities for model refinement. The study highlights flavonoid scaffolds with optimized LogP (2–3) and TPSA (<80 Ų) as high-priority candidates for synthesis. By integrating QSAR with stability-driven descriptor selection, this work provides a computational roadmap for accelerating flavonoid-based melanoma drug discovery. Future directions include experimental validation of top-ranked derivatives and incorporation of nonlinear machine learning methods to capture complex structure-activity landscapes.