PepGraphormer: an ESM-GAT hybrid deep learning framework for antimicrobial peptide prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Traditional antimicrobial peptide (AMP) prediction methods struggle with complex sequence-function relationships.
• Large Language Models (LLMs) like ESM2 excel at extracting deep protein sequence features.
• Graph Neural Networks (GNNs), specifically Graph Attention Networks (GATs), capture inter-residue relationships.

Purpose of the Study:

• To propose PepGraphormer, a novel fusion model for enhanced AMP prediction.
• To integrate ESM2's semantic feature extraction with GAT's structural learning capabilities.
• To advance computational methods for therapeutic peptide discovery.

Main Methods:

• Constructing a heterogeneous graph with peptide sequences and amino acids as nodes.
• Leveraging ESM2 for initial node embeddings and direct predictions.
• Utilizing GATs to learn graph-based representations and capture compositional patterns.
• Fusing ESM2 and GAT predictions for final classification.

Main Results:

• PepGraphormer significantly outperforms state-of-the-art models on multiple AMP prediction datasets.
• The model demonstrates excellent accuracy and stability in predicting antimicrobial peptides.
• Ablation and generalization experiments confirm the framework's effectiveness and robustness.

Conclusions:

• PepGraphormer offers a powerful new approach for AMP prediction by combining LLMs and GATs.
• The model effectively captures complex sequence-function relationships without requiring 3D structural information.
• This fusion framework presents a promising avenue for accelerating the discovery of novel therapeutic peptides.