Predicting Antimicrobial Resistance and Associated Genomic Features from Whole-Genome Sequencing

Thanks to the genomics revolution, thousands of strain-specific whole-genome sequences are now accessible for a wide range of pathogenic bacteria. This availability enables big data informatics approaches to be used to study the spread and acquisition of antimicrobial resistance (AMR). In this issue of the Journal of Clinical Microbiology, Nguyen et al. (M. Nguyen, S. W. Long, P. F. McDermott, R. J. Olsen, R. Olson, R. L. Stevens, G. H. Tyson, S. Zhao, and J. J. Davis, J Clin Microbiol 57:e01260-18, 2019, https://doi.org/10.1128/JCM.01260-18) report the results obtained with their machine learning models based on whole-genome sequencing data to predict the MICs of antibiotics for 5,728 nontyphoidal Salmonella genomes collected over 15 years in the United States. Their major finding demonstrates that MICs can be predicted with an average accuracy of 95% within ±1 2-fold dilution step (confidence interval, 95% to 95%), an average very major error rate of 2.7%, and an average major error rate of 0.1%. Importantly, these models predict MICs with no a priori information about the underlying gene content or resistance phenotypes of the strains, enabling the possibility to identify AMR determinants and rapidly diagnose and prioritize antibiotic use directly from the organism sequence. Employing such tools to diagnose and limit the spread of resistance-conferring mechanisms could help ameliorate the looming antibiotic resistance crisis.