Versatile Framework for Drug-Target Interaction Prediction by Considering Domain-Specific Features.

Area of Science:

• Computational Biology
• Drug Discovery
• Machine Learning

Background:

• Predicting drug-target interactions (DTIs) is vital for drug discovery but is hindered by costly experiments.
• Deep learning models show promise for DTI prediction but often fail to generalize to novel drug-target pairs due to limited training data.

Purpose of the Study:

• To develop a novel strategy, E-DIS (Ensemble of Deep learning models integrating Domain-specific features), to improve the generalization ability of DTI prediction models.
• To enhance the representation of proteins and ligands by capturing diverse domain-generic and domain-specific features.

Main Methods:

• An ensemble of models (E-DIS) was trained to capture both domain-generic and domain-specific features.
• Multiple 'expert' models were trained on different domains to learn and align domain-specific information without accessing unseen domain data.
• The approach aimed to adapt models to out-of-distribution data.

Main Results:

• E-DIS demonstrated significantly improved model performance in both in-domain and cross-domain settings across four benchmark datasets.
• The method showed superior domain generalization capabilities compared to existing DTI prediction techniques.
• E-DIS provided a comprehensive representation of proteins and ligands by effectively capturing diverse features.

Conclusions:

• The E-DIS strategy represents a significant advancement in DTI prediction by effectively combining domain-generic and domain-specific features.
• This approach enhances the generalization ability of DTI prediction models, paving the way for more efficient drug discovery.
• E-DIS offers a robust solution for predicting drug-target interactions, particularly for novel or unseen drug-target pairs.