Drug-target interaction prediction with tree-ensemble learning and output space reconstruction

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Drug-target interaction (DTI) prediction is crucial but experimentally challenging and costly.
• Existing machine learning methods for DTI prediction face accuracy and efficiency limitations.

Purpose of the Study:

• To develop a novel, accurate, and efficient in silico method for DTI prediction.
• To address DTI prediction as a multi-output prediction task integrating diverse background information.

Main Methods:

• Proposed a new learning method using ensembles of multi-output bi-clustering trees (eBICT) on reconstructed networks.
• Modeled DTI networks where nodes (drugs, proteins) are represented by features and interactions form the output space.
• Integrated background information from both drug and target protein spaces into a global network framework.

Main Results:

• Empirically evaluated the proposed approach against state-of-the-art DTI prediction methods.
• Demonstrated the effectiveness of the approach across various prediction settings using benchmark datasets.
• Showcased that output space reconstruction significantly boosts the predictive performance of tree-ensemble learning methods for more accurate DTI predictions.

Conclusions:

• Introduced a novel DTI prediction method utilizing bi-clustering trees on reconstructed networks.
• Building tree-ensemble models with output space reconstruction yields superior prediction results.
• The method preserves the advantages of tree-ensembles, including scalability, interpretability, and inductive capabilities.