Efficient Classification and Regression Models for the QSAR of Chloroquine Analogues against Chloroquine-Sensitive and Chloroquine-Resistant Plasmodium falciparum
Letters in Applied NanoBioScience · 2024
The treatment of Plasmodium falciparum malaria using chloroquine (CQ) has been hampered by resistance, and there is a need to continually optimize CQ for a more effective analog by structural modification using the QSAR approach. The study developed classification and regression-based models for the optimization and prediction of activities of CQ derivatives against CQ-sensitive (PfD-10) and CQ-resistant (PfW-2) P. falciparum strains using a comparative molecular field analysis (CoMFA) and machine learning (ML) algorithms. This present work involves extracting the molecular features and SMILES of CQ analogs from RDKit software and combined to form X- and Y-matrices. The classification model was trained using Python and evaluated using the confusion matrix, accuracy, precision, recall, and F1-score. The CoMFA model was constructed with conformers of the training set (n=20) by the PLS method, cross-validated by leave-one-out, and externally predicted the IC50 of the test set (n=7) using Open3dqsar of the Open3dtools. The calibration R2, cross-validation, Q2, and prediction P2 for the y-scrambled data set were 0.996 - 0.684 and 0.998 - 0.682 for PfD-10 and PfW-2 respectively. The decision tree classifier and Naïve Bayes algorithms outperformed the others in the classification model with accuracy scores of ≥ 0.88 against PfD-10 and PfW-2. A single ensemble model obtained by staking all the classification models significantly improved overall accuracy and reliability. The study developed models for the prediction of activity and provided an explanation for the SARs and strategies for further optimization of CQ analogs against PfD-10 and PfW-2 strains.