In silico QSAR and design of chalcone derivatives for HT-29 colorectal cancer: MLR and ANN approaches

Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide, necessitating the continuous discovery of potent and selective therapeutic agents. Chalcone derivatives have demonstrated significant cytotoxic potential against the HT-29 colorectal cancer cell line. This study aimed to develop robust Quantitative Structure-Activity Relationship (QSAR) models to predict anticancer activity and design novel chalcone derivatives by comparing Multiple Linear Regression (MLR) and Artificial Neural Network (ANN) approaches. A dataset comprising 193 chalcone derivatives was analyzed using 2D molecular descriptors. Model reliability was rigorously evaluated through internal validation (LOO and LMO) and external cross-validation (Q_F1^2, Q_F2^2, Q_F3^2). The results demonstrated that the Stepwise MLR model (27 descriptors) outperformed the ANN approach, exhibiting superior stability and predictive power with R2 = 0.817, Q_LOO^2= 0.744, and RMSEP = 0.217. In contrast, the ANN model (13i-8N-1O architecture) showed clear indications of overfitting with a negative Q_LMO^2 of -1.957. The most influential descriptors identified were QCmin (+1.173), MATSv2 (+1.043), and UI (-0.806). Based on the optimized model, a novel lead compound, Modifikasi_W_136a, was designed with chloro, fluoro, and trifluoromethoxy substitutions, achieving a predicted pIC50 of 7.04. An in silico toxicity assessment using ProTox-III revealed a Class 4 acute toxicity profile with favorable hepatotoxicity and genotoxicity predictions, though specific alerts for nephrotoxicity and cardiotoxicity were identified requiring experimental follow-up. This study provides a validated computational framework for the rational design of colorectal anticancer agents with integrated safety profiling.