GSF-DTA: An Innovative Graph-Sequence Fusion Framework for Drug-Target Affinity Prediction.

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Drug development relies heavily on predicting drug-target affinity (DTA).
• Experimental DTA prediction methods are accurate but costly and slow.
• Computational methods offer scalability but often use limited data types (sequence or graph).

Purpose of the Study:

• To develop a novel computational framework, GSF-DTA, for improved DTA prediction.
• To integrate both graph-based structural features and sequence-derived representations for a comprehensive interaction analysis.
• To enhance the accuracy and generalizability of DTA prediction models.

Main Methods:

• Developed GSF-DTA, a graph-sequence fusion framework.
• Integrated graph-based drug features and protein sequence-based target features.
• Validated the model on the large-scale BindingDB dataset.

Main Results:

• GSF-DTA achieved superior predictive accuracy compared to existing methods.
• The model demonstrated strong generalization capabilities on the BindingDB dataset.
• GSF-DTA showed robust performance in cold-start scenarios for predicting novel drug-target interactions.

Conclusions:

• GSF-DTA offers a promising and generalizable strategy for DTA prediction.
• The fusion of graph and sequence data improves understanding of drug-target interactions.
• This approach can accelerate drug design and discovery processes.