Predictive classifier models built from natural products with antimalarial bioactivity using machine learning...

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and natural products research
• Machine learning in pharmacology

Background:

• Natural products are a rich source of potential antimalarial compounds.
• Experimental bioassays for antiplasmodial activity are costly and time-consuming.
• Predictive modeling can prioritize natural products for further investigation.

Purpose of the Study:

• To develop accurate machine learning models for predicting the antiplasmodial bioactivity of natural products.
• To identify potential antimalarial drug candidates from large chemical libraries.
• To elucidate key chemical features associated with antiplasmodial activity.

Main Methods:

• Utilized classical machine learning algorithms including Naïve Bayesian, Voted Perceptron, Random Forest, and Sequential Minimization Optimization of Support Vector Machines.
• Trained models using in-vitro antiplasmodial activity data, molecular descriptors, and two-dimensional molecular fingerprints.
• Evaluated model performance using an independent test dataset and analyzed chemical features linked to activity.

Main Results:

• Random Forest and Sequential Minimization Optimization models demonstrated strong predictive performance.
• Random Forest achieved 82.81% accuracy, while SMO reached 85.93% accuracy.
• The amine chemical group, specifically alkyl amines and basic nitrogen, was identified as crucial for antiplasmodial activity.

Conclusions:

• Machine learning models can effectively predict the antiplasmodial bioactivity class of natural products.
• This approach aids in the efficient screening of natural products for antimalarial potential.
• Identifying essential chemical features like amine groups can guide the design of new antimalarial compounds.