Development and rigorous validation of antimalarial predictive models using machine learning approaches.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and development
• Parasitology and infectious diseases

Background:

• The ChEMBL database contains a large collection of experimentally verified compounds.
• Antimalarial drug discovery is crucial for combating Plasmodium falciparum infections.

Purpose of the Study:

• To build and evaluate machine learning models for predicting antimalarial activity.
• To identify robust models for facilitating the discovery of new antimalarial agents.

Main Methods:

• Utilized support vector machine (SVM), random forest (RF), k-nearest neighbour (kNN), and XGBoost algorithms.
• Employed a feature selection framework to optimize model descriptors.
• Evaluated model performance using applicability domain, Y-scrambling, AUC-ROC curves, probability calibration, and predictiveness curves.

Main Results:

• SVM and XGBoost models demonstrated the highest performance, achieving ~85% accuracy on an independent test set.
• Both SVM and XGBoost models exhibited good probability calibration.
• Predictiveness curves showed high total gain for SVM (0.67) and XGBoost (0.75).
• Models successfully predicted high-affinity compounds from an external PubChem antimalarial bioassay.

Conclusions:

• The developed SVM and XGBoost models are robust and reliable for predicting antimalarial activity.
• These models can significantly aid in the virtual screening and discovery of novel antimalarial compounds.
• The study highlights the potential of machine learning in accelerating drug discovery efforts against malaria.