QSAR machine learning-guided discovery of renieramycin right-half derivatives with cytotoxic activity against NSCLC cells

This study explores the structure-activity relationships of renieramycin right-half and full-skeleton compounds using quantitative structure-activity relationship (QSAR) modeling. Linear (Genetic Algorithm-Multiple Linear Regression, GA-MLR) and non-linear machine learning approaches (Random Forest, Support Vector Regression, and XGBoost) were employed to develop predictive models with quantum chemical and molecular descriptors. The best-performing model, using Support Vector Regressor (SVR), achieved a coefficient of determination (R²) of 0.946 for the training set and a root mean square error (RMSE) of 0.246 for the test set. Key descriptors influencing cytotoxicity included charges at C2 and C4, HOMO energy, and polarizability. External validation with newly synthesized renieramycin right-half derivatives yielded an RMSE of 0.236. The machine learning-based QSAR models demonstrated exceptional accuracy and reliability in cytotoxicity prediction, underscoring their utility in guiding the design of novel renieramycin derivatives. The cytotoxicity test of the newly synthesized renieramycin shows an anomaly than the previous experimental findings, i.e., the O-benzyl containing derivative was more cytotoxic than the hydroxyl or quinone containing renieramycin derivatives. These findings highlight the potential of fine-tuned QSAR methodologies to accelerate the development of highly effective anticancer agents based on renieramycin right-half structures.