SGNet: Sequence-Based Convolution and Ligand Graph Network for Protein Binding Affinity Prediction.

Area of Science:

• Computational chemistry and structural biology
• Bioinformatics and cheminformatics

Background:

• Accurate prediction of protein-ligand binding affinity is vital across scientific fields.
• Existing computational methods often require solved protein structures, limiting their application.
• A significant number of proteins lack determined 3D structures, posing a challenge for binding affinity prediction.

Purpose of the Study:

• To develop a novel computational method for predicting protein-ligand binding affinity using only amino acid sequences and ligand molecular graphs.
• To integrate protein sequence information with ligand molecular structure information for enhanced prediction accuracy.
• To address the limitations of structure-dependent methods in binding affinity prediction.

Main Methods:

• Proposed a sequence-based convolution and ligand graph network (SGNet).
• Utilized Conjoint Triad (CT) encoding and a 1D convolutional neural network (CNN) for protein sequence feature extraction.
• Employed a graph attention network (GAT) for ligand molecular feature extraction.
• Fused protein and ligand features for binding affinity prediction via a fully connected layer.

Main Results:

• SGNet demonstrated strong prediction performance on both KIKD and IC50 datasets.
• Achieved low root-mean-square errors (RMSE) of 1.287 (KIKD) and 1.58 (IC50).
• Obtained high correlation coefficients (Pearson R) of 0.687 (KIKD) and 0.592 (IC50).
• Outperformed established methods like Kdeep and GraphDTA in comparative experiments.

Conclusions:

• SGNet effectively predicts protein-ligand binding affinity by integrating sequence and graph-based molecular information.
• The method overcomes the need for solved protein structures, broadening applicability in drug discovery and molecular design.
• SGNet represents a significant advancement in sequence-based computational approaches for predicting molecular interactions.