A Robust Regression-Based Modeling to Predict Antiplasmodial Activity of Thiazolyl-pyrimidine hybrid derivatives against Plasmodium falciparum

Thiazolyl-pyrimidine hybrid plays significant roles in the biological activities and SAR of thiazolylpyrimidines (Tzpd), thiazolopyrimidines and thienopyrimidines due the combination of the thiazole and pyrimidine pharmacophores. The study developed regression-based models for the prediction of antiplasmodial activity of 43 Tzpd hybrid obtained from the ChEMBL database. The molecular descriptors (145 features) were scaled down to 6 using the recursive feature elimination. The X- and Y-matrix were split into 34 train and 9 test sets using a split ratio of 0.20. Regression models were built using scikit-learn algorithms: k-Nearest Neighbours (kNN), Support Vector Regressor (SVR) and Random Forest Regressor (RFR)) to predict the pIC₅₀ of the test set. The models were evaluated using R², mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), p-values, F-statistic, and variance inflation factor (VIF). Of the 145 features calculated for the 43 TzPd, 6 molecular features: FCASA-, MNDO_LUMO, E_str, vsurf_HB1, vsurf_G and vsurf_DD12 (p < 0.05; VIF < 5) were found to significantly influence the antiplasmodial activity. Five-fold cross-validation performance scores of kNN, SVR, and RFR showed that the performance metrics of SVR (MSE = 0.052; R² = 0.610; MAE = 0.174; RMSE = 0.230; pIC₅₀(predicted) = 8.06 - 0.45vsurf_G + 0.37FCASA‒ - 0.42MNDO_LUMO – 0.20E_str + 0.30vsurf_HB1 – 0.38vsurf_DD12) outperformed models. The study developed predictive models and provided insights into the chemical features necessary for the optimization of thiazolyl-pyrimidine to enhance antiplasmodial activity.