Metagenomic Antimicrobial Susceptibility Testing from Simulated Native Patient Samples.

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Genomic antimicrobial susceptibility testing (AST) is accurate for isolates but not systematically evaluated for clinical metagenomic data.
• Antimicrobial resistance (AMR) is a growing global health threat, necessitating novel diagnostic approaches.

Purpose of the Study:

• To evaluate the performance of in-silico genomic AST from clinical metagenomes (MG-AST).
• To investigate factors influencing MG-AST accuracy, including sequencing, complexity, and bioinformatics.
• To compare rule-based and machine learning approaches for MG-AST.

Main Methods:

• Simulated over 2000 complicated urinary tract infection (cUTI) metagenomes using isolate sequencing data and septic urine samples.
• Applied rule-based and machine learning-based genomic AST classifiers.
• Optimized metagenomics assembly, binning, and plasmid contig reassignment strategies.

Main Results:

• MG-AST achieved balanced accuracy within 5.1% of isolate-derived genomic AST using an optimized workflow.
• Metagenome complexity and taxonomic relatedness impacted binning and MG-AST accuracy in polymicrobial samples.
• Reassigning plasmid contigs and analyzing whole resistomes improved performance on complex samples.
• Machine learning-based MG-AST demonstrated superior accuracy over rule-based methods on simulated patient samples.

Conclusions:

• Optimized MG-AST workflows provide accurate AMR predictions from clinical metagenomic data.
• Machine learning approaches enhance MG-AST accuracy, particularly for complex polymicrobial infections.
• MG-AST holds promise for rapid, culture-independent AMR diagnostics in clinical settings.