GS-DTI: a graph-structure-aware framework leveraging large language models for drug-target interaction prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate prediction of drug-target interactions (DTIs) is crucial for drug discovery, especially for novel targets and compounds.
• Graph neural networks and pre-trained models offer advanced capabilities for capturing molecular features and improving DTI prediction generalization.

Purpose of the Study:

• To develop a robust and generalizable framework for predicting drug-target interactions (DTIs).
• To integrate molecular graph transformers, protein language models, and protein tertiary structure for enhanced DTI prediction.
• To provide interpretable predictions and improve performance in challenging cross-domain settings.

Main Methods:

• GS-DTI framework utilizes molecular graph transformers for drug feature extraction from SMILES.
• Protein features are derived from both sequence and predicted 3D structure using protein language models.
• A multi-task loss function with contrastive learning enhances generalization and interpretability.

Main Results:

• GS-DTI achieves state-of-the-art performance on benchmark datasets and cross-domain settings.
• The model shows over 10% improvement in Matthews Correlation Coefficient (MCC) for drug-target pair cold-start prediction.
• GS-DTI accurately identifies binding pockets, offering robust interpretability and demonstrating potential in virtual screening.

Conclusions:

• GS-DTI provides a powerful and interpretable approach for DTI prediction.
• The framework demonstrates significant improvements in generalization, particularly for novel targets and compounds.
• GS-DTI holds promise for accelerating drug discovery through efficient virtual screening.