GraphPLBR: Protein-Ligand Binding Residue Prediction With Deep Graph Convolution Network.

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Protein-ligand interactions are mediated by specific amino acid residues.
• Identifying these binding residues is vital for understanding protein function and drug design.
• Experimental methods for residue identification are time-consuming, necessitating computational approaches.

Purpose of the Study:

• To develop an accurate and efficient computational method for predicting protein-ligand binding residues (PLBR).
• To introduce GraphPLBR, a novel framework utilizing Graph Convolutional Neural Networks (GCNs) for PLBR prediction.

Main Methods:

• Representing proteins as graphs with residues as nodes, transforming PLBR prediction into a graph node classification task.
• Employing deep GCNs to extract features from higher-order neighbors.
• Implementing initial residue connection with identity mapping to mitigate over-smoothing issues in deep GCNs.

Main Results:

• GraphPLBR demonstrates superior performance compared to existing state-of-the-art methods on several evaluation metrics.
• The graph node classification approach offers a unique and innovative perspective for PLBR prediction.

Conclusions:

• GraphPLBR provides an effective computational solution for identifying protein-ligand binding residues.
• This GCN-based framework advances the field of computational drug design and protein function interpretation.