AMR-GNN: a multi-representation graph neural network framework to enable genomic antimicrobial resistance prediction.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Whole-genome sequencing (WGS) provides rich data for understanding antimicrobial resistance (AMR).
• High dimensionality and lack of standardized genomic representations hinder accurate AMR phenotype prediction from WGS data.
• Existing machine learning (ML) approaches face challenges in leveraging complex genomic information for AMR prediction.

Purpose of the Study:

• To develop and validate AMR-GNN, a novel graph deep learning framework for AMR phenotype prediction using WGS data.
• To integrate multiple genomic representations within a graph neural network (GNN) architecture to enhance prediction accuracy.
• To address limitations in ML-based AMR prediction, including improving performance, mitigating clonal effects, and identifying predictive biomarkers.

Main Methods:

• Developed AMR-GNN, a graph deep learning framework utilizing graph neural networks (GNNs).
• Integrated multiple genomic representations to capture diverse genetic features relevant to AMR.
• Applied the framework to Pseudomonas aeruginosa and validated on a large dataset of Gram-negative and Gram-positive pathogens.

Main Results:

• AMR-GNN demonstrated enhanced performance in AMR phenotype prediction compared to traditional ML methods.
• The framework successfully mitigated the influence of clonal relationships on prediction accuracy.
• Identification of informative genomic biomarkers for AMR was achieved, providing explainability for predictions.
• Broad applicability was confirmed across diverse pathogen types and drug combinations.

Conclusions:

• AMR-GNN offers a powerful, data-driven approach for accurate AMR phenotype prediction from WGS data.
• The framework's ability to integrate multiple genomic features and provide explainability represents a significant advancement in the field.
• AMR-GNN shows promise for widespread application in clinical microbiology and infectious disease research.