Predicting Antimalarial Activity in Natural Products Using Pretrained Bidirectional Encoder Representations from...

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Malaria remains a significant global health threat with high annual prevalence.
• Research actively seeks effective antimalarial compounds from both synthetic and natural sources.
• Computational methods offer a promising avenue for identifying natural product antimalarials.

Purpose of the Study:

• To develop a novel computational approach for identifying natural antimalarial products.
• To evaluate the efficacy of a new molecular encoding scheme based on Bidirectional Encoder Representations from Transformers (BERT).
• To compare the performance of machine learning models utilizing this novel encoding scheme against existing methods.

Main Methods:

• A novel molecular encoding scheme, NPBERT, was developed using Bidirectional Encoder Representations from Transformers.
• Four machine learning algorithms were employed: k-Nearest Neighbors (k-NN), Support Vector Machines (SVM), eXtreme Gradient Boosting (XGB), and Random Forest (RF).
• Prediction models were constructed to identify natural products with antimalarial properties.

Main Results:

• Support Vector Machines (SVM) demonstrated the highest performance among the tested classifiers, followed by XGBoost, k-NN, and Random Forest.
• The proposed NPBERT molecular encoding scheme showed superior effectiveness compared to existing state-of-the-art methods.
• The study validated the potential of transformer-based models in molecular encoding for drug discovery.

Conclusions:

• NPBERT, a transformer-based molecular encoder, is highly effective for identifying natural antimalarial products.
• Machine learning models, particularly SVM, integrated with NPBERT, offer a powerful tool for drug discovery.
• The application of transformer architectures in molecular encoding has broad potential for various biomedical applications.