Performance and hypothetical clinical impact of an mNGS-based machine learning model for antimicrobial susceptibility...

Antimicrobial resistance is an escalating global health crisis, underscoring the urgent need for timely and targeted therapies to ensure effective clinical treatment. We developed a machine learning model based on metagenomic next-generation sequencing (mNGS) for rapid antimicrobial susceptibility prediction (mNGS-based AST), which was tailored to five ESKAPEE bacteria: Acinetobacter baumannii, Klebsiella pneumoniae, Escherichia coli, Pseudomonas aeruginosa, and Staphylococcus aureus. However, the clinical utility remained unvalidated. Assuming that mNGS-based AST results were obtained during clinical management, we assessed its clinical utility using data from a previous observational cohort study of clinical mNGS applications. We collected the data from 114 patients infected with five ESKAPEE bacteria from 07/2021 to 03/2023 and incorporated the sequencing data into the model. We evaluated the performance and hypothetical impact of the method by comparing its results and therapy recommendations with those based on traditional culture-based AST. The primary outcome was the performance of mNGS-based AST (n = 113 strains). mNGS-based AST displayed an overall accuracy of 93.84% and shorter turnaround time (1.12 ± 0.33 days vs 2.81 ± 0.57 days for culture-based AST, t = -27.31, P < 0.05). The secondary outcomes included the proportion of patients who could benefit from mNGS-based AST. It could allow earlier and suitable antibacterial adjustments in 32.05% of culture-positive patients (25/78) and offer actionable antimicrobial susceptibility results in 16.67% of culture-negative cases (6/36). mNGS-based AST offers a promising approach for individualized antibacterial therapy.

Metagenomic next-generation sequencing (mNGS)-based antimicrobial susceptibility prediction (AST) is a novel method for predicting the antimicrobial susceptibility of ESKAPEE bacteria using a machine learning approach and short-read sequencing data. Assuming that mNGS-based AST results were obtained during clinical management, it could significantly reduce turnaround time while maintaining a high level of accuracy, allowing for earlier therapeutic adjustments for patients. Furthermore, mNGS-based AST can be integrated with clinical mNGS to maximize the utility of short-read data without substantial cost increases. This study demonstrates the potential of mNGS-based AST for precise, individualized antibacterial selection and highlights its broader applicability in enhancing clinical antimicrobial use for various infections.