SGAC: a graph neural network framework for imbalanced and structure-aware AMP classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) are crucial for combating antibiotic resistance.
• Existing AMP classification methods often neglect spatial structural information, limiting accuracy.
• Current graph-based methods can be overly complex due to atom-level details.

Purpose of the Study:

• To develop a novel framework for accurate AMP classification using peptide 3D structures.
• To address the challenge of class imbalance in AMP identification.
• To improve the capture of spatial topology and backbone geometry for peptide representation.

Main Methods:

• Utilized OmegaFold for lightweight prediction of peptide 3D structures.
• Constructed peptide graphs using Cα atoms to represent backbone geometry.
• Developed a spatial graph neural network (GNN)-based classifier (SGAC).
• Incorporated weight-enhanced contrastive learning and pseudo-label distillation to handle class imbalance.

Main Results:

• SGAC achieved state-of-the-art performance on AMP and non-AMP datasets.
• The method demonstrated superior accuracy compared to existing baseline approaches.
• Successfully balanced representation learning despite significant class imbalance.

Conclusions:

• SGAC offers a powerful new approach for classifying antimicrobial peptides by integrating structural information.
• The framework effectively overcomes limitations of sequence-based methods and complex graph constructions.
• This work provides a valuable tool for discovering novel AMPs to combat antibiotic resistance.