Predicting Antimicrobial Peptides Using ESMFold-Predicted Structures and ESM-2-Based Amino Acid Features with Graph...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Antimicrobial resistance (AMR) is a global health crisis, necessitating novel therapeutic strategies.
• Antimicrobial peptides (AMPs) show promise as alternatives to conventional antibiotics.
• Current deep learning models for AMP prediction often rely on amino acid sequences and can be computationally intensive.

Purpose of the Study:

• To develop an alignment-free framework for predicting antimicrobial peptides (AMPs) using peptide 3D structures and evolutionary information.
• To assess the performance of graph attention networks (GATs) within this framework for AMP classification.
• To improve upon existing state-of-the-art methods in terms of accuracy, speed, and memory efficiency.

Main Methods:

• A novel framework, esm-AxP-GDL, was developed to generate graph representations from ESMFold-predicted peptide 3D structures.
• Amino acid-level evolutionary information from Evolutionary Scale Modeling (ESM-2) was incorporated into the graph nodes.
• Graph attention networks (GATs) were trained and evaluated on a dataset of 67,058 peptides for AMP classification.

Main Results:

• The proposed GAT models demonstrated superior generalization abilities compared to 20 existing non-DL and DL-based models.
• Optimal models were achieved using evolutionary data from 36- and 33-layer ESM-2.
• Fusion of best-performing GAT models, which captured distinct chemical spaces, significantly enhanced classification accuracy.

Conclusions:

• The esm-AxP-GDL framework provides an effective, structure-dependent, and alignment-free approach for AMP screening.
• This methodology offers a promising tool for classifying AMPs and potentially other peptide/protein activities.
• The framework addresses limitations of previous methods by integrating 3D structural information with evolutionary data efficiently.