NegStacking: Drug-Target Interaction Prediction Based on Ensemble Learning and Logistic Regression.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-target interactions (DTIs) are vital in drug research.
• Predicting DTIs using machine learning often faces challenges due to imbalanced datasets (fewer positive than negative samples).
• Existing methods inadequately leverage large negative sample sets, limiting prediction accuracy.

Purpose of the Study:

• To propose a stacking framework, NegStacking, to improve DTI prediction performance.
• To address the insufficient utilization of negative samples in DTI prediction models.
• To enhance the efficiency of the drug discovery process through better DTI identification.

Main Methods:

• Developed a stacking framework (NegStacking) for DTI prediction.
• Employed sampling to generate diverse negative sample sets.
• Trained weak learners on different negative and common positive sample sets.
• Utilized logistic regression (LR) as a meta-learner for adaptive combination.
• Incorporated feature subspacing and hyperparameter perturbation to boost ensemble diversity.

Main Results:

• NegStacking demonstrated superior performance compared to existing DTI prediction methods.
• The framework effectively improves the prediction of drug-target interactions.
• Experimental results validate the model's effectiveness and potential.

Conclusions:

• NegStacking offers a significant advancement in DTI prediction by addressing data imbalance.
• The proposed method shows broad application prospects for accelerating drug discovery.
• The source code and datasets are publicly available for further research and development.