Drug-target affinity prediction with extended graph learning-convolutional networks.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Computer-aided drug design (CADD) utilizes high-performance computing for pharmaceutical research.
• Drug-target affinity (DTA) prediction accelerates compound screening, reducing costs and resource use.
• Deep learning enhances DTA prediction accuracy, with graph-based methods showing promise for comprehensive data representation.

Purpose of the Study:

• To address limitations of fixed adjacent matrices in graph-based DTA prediction.
• To develop a model capable of learning richer structural information from protein and drug molecular graphs.
• To improve the generalization capabilities of DTA prediction models.

Main Methods:

• Introduction of GLCN-DTA, a model integrating a graph learning module with graph convolution.
• Learning a soft adjacent matrix to refine contextual structures of molecular graphs.
• Utilizing graph convolution for enhanced feature representation from protein and drug structures.

Main Results:

• GLCN-DTA demonstrates superior robustness and accuracy in DTA prediction tasks.
• The model effectively learns richer structural information compared to conventional fixed adjacent matrix approaches.
• Experimental validation confirms the efficacy of GLCN-DTA across diverse scenarios.

Conclusions:

• GLCN-DTA enhances DTA prediction by synergizing graph learning and graph convolution for richer representations.
• The model focuses on improving feature representation without differentiating protein classifications.
• GLCN-DTA shows potential effectiveness for structurally ordered proteins, with possible limitations for intrinsically disordered proteins.