Predicting Antimicrobial Resistance Using Partial Genome Alignments.

Area of Science:

• Genomics and Bioinformatics
• Microbiology
• Machine Learning in Health

Background:

• Antimicrobial resistance (AMR) poses a significant global health challenge, necessitating improved detection and prediction methods.
• Existing bioinformatic tools for AMR prediction often rely on whole-genome sequences or known AMR genes.
• Recent findings suggest that incomplete genomic data may suffice for accurate AMR phenotype prediction.

Purpose of the Study:

• To systematically investigate the predictive power of small genomic regions for AMR phenotypes using machine learning.
• To build and evaluate random forest classifiers for predicting AMR in Klebsiella pneumoniae, Mycobacterium tuberculosis, and Salmonella enterica.
• To identify specific genomic regions, including non-coding areas, associated with AMR.

Main Methods:

• Development of random forest-based machine learning classifiers.
• Training models on chromosomal alignments and progressively subsampled subalignments for K. pneumoniae, M. tuberculosis, and S. enterica.
• Analysis of predictive signals within identified subalignments to pinpoint genes and intergenic regions linked to AMR.

Main Results:

• Highly accurate prediction (≥70%) of AMR phenotypes was achieved using very small genomic regions (0.1-0.2% of the chromosome).
• Predictive subalignments often contained genes not previously associated with AMR, including those involved in virulence, transport, and stress survival.
• Identified specific chromosomal regions with high and low predictive signals for AMR across the studied bacterial species.

Conclusions:

• Small, conserved genomic regions are sufficient for predicting AMR phenotypes with high accuracy.
• The study implicates novel genes and intergenic regions in AMR, expanding our understanding beyond known resistance genes.
• These findings can inform the development of new diagnostic tools and mitigation strategies for combating antimicrobial resistance.