Quantitative structure-activity relationship (QSAR) studies of dihydroorotate dehydrogenase inhibitors of Plasmodium falciparum for the combat of malaria resistance using machine learning models classification approach

Ring-stage resistance to the first-line artemisinin-based antimalarial medication and the emergence of Plasmodium species resistance has reduced the efficacy of the majority of antimalarial medications. Compounds that bind tightly to Plasmodium falciparum dihydroorotate dehydrogenase ( Pf DHODH) inhibit parasite proliferation and disease. Pf DHODH inhibitors' IC50 values were taken from the ChEMBL database (ChEMBL ID CHEMBL3486). To ascertain which model had the highest performance, robustness, and interpretability, 12 machine learning models were built from 12 sets of chemical fingerprints using a final set of 465 inhibitors after curation. The datasets were separated into two categories: balanced and imbalanced. The balanced data set was subjected to both undersampling and oversampling techniques. From the findings, the balance oversampling technique gave the best outcome, with the majority of MCC train values above 0.8 and the majority of MCC CV and MCC tes t values exceeding 0.65. Because of its capacity to identify pertinent characteristics and its ease of understanding, we chose Random Forest (RF) over the other machine learning techniques. With >80% accuracy, sensitivity, and specificity in the internal set, cross-validation, and external sets, the SubstructureCount fingerprint provides the best overall. SubstructureCount MCC values in the external set, cross-validation, and training internal sets are 0.76, 0.78, and 0.97, respectively and with AMCC value of <2. The Gini index was used to assess the model's feature significance, which showed that Pf DHODH inhibitory activity was influenced by nitrogenous, fluorine, and oxygenation characteristics in addition to aromatic moieties and Chirality. The results of this investigation might expedite the drug development of Pf DHODH inhibitors by offering guidance for the future optimization of leads.