DNN-DTIs: Improved drug-target interactions prediction using XGBoost feature selection and deep neural network

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are crucial for understanding drug mechanisms, drug repositioning, and drug design.
• Experimental methods for DTI analysis are time-consuming and labor-intensive.
• Machine learning (ML) offers a more efficient alternative for DTI prediction.

Purpose of the Study:

• To propose a novel pipeline, DNN-DTIs, for accurate prediction of drug-target interactions.
• To leverage advanced machine learning techniques for improved DTI analysis.
• To facilitate drug repositioning and design through enhanced prediction capabilities.

Main Methods:

• Target proteins characterized by diverse features (e.g., pseudo-amino acid composition, structural features).
• Drug compounds encoded using substructure fingerprints.
• Feature selection using eXtreme gradient boosting (XGBoost) and data balancing with synthetic minority oversampling technique (SMOTE).
• Deep neural network (DNN) model implemented for DTI prediction.

Main Results:

• DNN-DTIs demonstrated superior performance compared to existing state-of-the-art predictors.
• Achieved high accuracy (ACC) values across multiple datasets: Enzyme (98.78%), Ion Channels (98.60%), GPCR (97.98%), Nuclear Receptors (98.24%), and Kuang's dataset (98.00%).

Conclusions:

• DNN-DTIs is a highly accurate and efficient tool for predicting drug-target interactions.
• The model's performance makes it valuable for drug repositioning and drug design initiatives.
• This approach offers a promising computational strategy for advancing pharmaceutical research.