Drug-target interaction prediction using ensemble learning and dimensionality reduction.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Experimental drug-target interaction (DTI) prediction is costly and time-consuming.
• Increasing data complexity challenges existing computational DTI prediction methods.
• Need for efficient computational approaches to predict drug-target interactions.

Purpose of the Study:

• To develop a novel framework for drug-target interaction prediction.
• To enhance prediction performance through feature dimensionality reduction and ensemble learning.
• To improve the efficiency and accuracy of computational drug-target interaction prediction.

Main Methods:

• Implemented feature subspacing to create diverse base learners for an ensemble model.
• Applied three distinct dimensionality reduction techniques to the feature subsets.
• Trained homogeneous base learners (Decision Tree, Kernel Ridge Regression) on reduced features.
• Aggregated base learner predictions to generate final drug-target interaction predictions.

Main Results:

• The EnsemKRR model achieved a high Area Under the ROC Curve (AUC) of 94.3% for drug-target interaction prediction.
• Dimensionality reduction positively impacted the performance of the EnsemDT model.
• The proposed framework demonstrated significant improvements over state-of-the-art methods.

Conclusions:

• The integrated approach of feature dimensionality reduction and ensemble learning effectively enhances drug-target interaction prediction.
• Ensemble Kernel Ridge Regression (EnsemKRR) shows superior performance in predicting drug-target interactions.
• The developed framework offers a more efficient and accurate computational solution for drug discovery.