Prediction of Drug-Target Interactions Based on Network Representation Learning and Ensemble Learning

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Drug-target protein interaction identification is crucial for drug development.
• Class imbalance exists due to fewer known interactions than unknown ones.
• Previous methods often neglect information from negative samples.

Purpose of the Study:

• To develop a novel method for predicting candidate drug-related proteins.
• To fully utilize negative samples to improve prediction performance.
• To address the class imbalance problem in drug-protein interaction prediction.

Main Methods:

• A novel method, NGDTP (Non-negative Matrix Factorisation and Gradient Boosting Decision Tree), was developed.
• NGDTP integrates multiple drug and protein similarities and interactions.
• Network representation learning using matrix factorization and a GBDT-based prediction model were employed.

Main Results:

• NGDTP fully utilizes all negative samples, effectively alleviating class imbalance.
• The method achieves superior prediction performance compared to state-of-the-art methods.
• NGDTP successfully retrieves more actual drug-protein interactions in top predictions.

Conclusions:

• NGDTP demonstrates superior performance in predicting drug-protein interactions.
• The method aids in identifying potential candidate proteins for drugs.
• NGDTP shows significant promise for accelerating drug discovery and development.