CGPDTA: An Explainable Transfer Learning-Based Predictor With Molecule Substructure Graph for Drug-Target Binding...

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-target interaction (DTI) identification is vital for drug discovery.
• Experimental determination of drug-target binding affinities (DTAs) is time-intensive.
• Current DTI prediction methods struggle with feature representation and interpretability.

Purpose of the Study:

• To develop a novel deep learning framework, CGPDTA, for accurate DTA prediction.
• To enhance feature representation and interpretability in drug-target interaction prediction.
• To leverage transfer learning and diverse data sources for improved DTI analysis.

Main Methods:

• CGPDTA utilizes transfer learning and advanced drug/protein language models.
• Incorporates molecular substructure graphs for drug representation.
• Employs protein pocket sequences for target representation.
• Integrates drug-drug and protein-protein interaction knowledge.

Main Results:

• CGPDTA demonstrates superior accuracy compared to existing DTI prediction methods.
• The framework provides enhanced interpretability of the prediction process.
• Effective capture of local drug and target features is achieved.

Conclusions:

• CGPDTA represents a significant advancement in predicting drug-target binding affinities.
• The model offers improved accuracy and interpretability for drug discovery pipelines.
• CGPDTA facilitates a deeper understanding of drug-target interactions.