MACHINE LEARNING-BASED QSAR MODELING OF TRICLOSAN ANALOGS FOR ANTIPLASMODIAL ACTIVITY PREDICTION USING SHRINKAGE REGRESSION MODEL AND TOPOLOGICAL-PHYSICOCHEMICAL DESCRIPTORS
TL;DR
The RFECV–Ridge QSAR framework provides a reliable and interpretable approach for virtual screening and rational design of novel pfENR inhibitors and highlights the importance of carefully selected descriptors in improving both interpretability and accuracy.
Abstract
Drug-resistant Plasmodium falciparum continues to threaten malaria control, necessitating new antimalarial discovery strategies. QSAR modeling with machine learning offers a cost-effective approach to relate molecular features to biological activity and prioritize candidate compounds for further development. In particular, topological descriptors (capturing molecular connectivity, branching, and structural arrangement) and physicochemical descriptors (such as hydrophobicity, electronic properties, polarity, and steric effects) provide complementary and mechanistically meaningful information governing ligand–target interactions and biological response. Here, a ridge regression QSAR model was developed to predict the anti-plasmodial activity of triclosan analogs. RFECV (Recursive Feature Elimination with Cross-Validation) was used for optimal descriptor selection. The model achieved an R² of 88.83% on the training set and an R²_test of 80.80% on the external test set, indicating strong predictive performance and robust generalization. These results highlight the importance of carefully selected descriptors in improving both interpretability and accuracy. Overall, the RFECV–Ridge QSAR framework provides a reliable and interpretable approach for virtual screening and rational design of novel pfENR inhibitors.