Pangenome-based interpretable machine learning framework for predicting antimicrobial resistance in foodborne...

Area of Science:

• Microbiology
• Genomics
• Computational Biology

Background:

• Antimicrobial resistance (AMR) in foodborne Escherichia coli (E. coli) poses a significant public health threat.
• Predicting AMR from whole-genome sequencing data is crucial for surveillance but conventional methods focusing on known resistance genes are limited.
• The broader genomic context associated with AMR is often overlooked in traditional approaches.

Purpose of the Study:

• To develop interpretable machine learning models for predicting AMR in foodborne E. coli using pangenome-derived accessory genes.
• To identify genomic features, beyond canonical resistance genes, that contribute to AMR prediction.
• To explore the role of mobile genetic elements and stress response mechanisms in AMR.

Main Methods:

• Utilized whole-genome sequencing data from 655 foodborne E. coli isolates.
• Developed interpretable machine learning models based on pangenome-derived accessory genes.
• Employed SHAP analysis to interpret model predictions and identify important genomic features.

Main Results:

• Achieved robust predictive performance for AMR across five antibiotics.
• Identified genomic features beyond known resistance genes, including mobile genetic element markers and stress response mechanisms.
• The disinfectant resistance gene qacEΔ1 was a consistently important predictive feature, suggesting its role as a positional marker for co-selection.

Conclusions:

• A pangenome-based framework enables interpretable AMR predictions in foodborne E. coli.
• Genomic context, including mobile genetic elements and specific genes like qacEΔ1, provides valuable signals for AMR surveillance.
• This approach can capture potential genomic signals linked to resistance dissemination in food production environments.