Predicting S. aureus antimicrobial resistance with interpretable genomic space maps

Abstract

AbstractIncreasing antimicrobial resistance (AMR) represents a global healthcare threat. Methods for rapid selection of optimal antibiotic treatment are urgently needed to decrease the spread of AMR and associated mortality. The use of machine learning (ML) techniques based on genomic data to predict resistance phenotypes serves as a solution for the acceleration of the clinical response prior to phenotypic testing. Nonetheless, many existing ML methods lack interpretability and do not implicitly incorporate visualization of the sequence space that can be useful for extracting insightful patterns from genomic data. Herein, we present a methodology for AMR prediction and visualization of sequence space based on the non-linear dimensionality reduction method □ generative topographic mapping (GTM). This approach applied to data on AMR of >5000 S. aureus isolates retrieved from the PATRIC database yielded GTM models with reasonable accuracy for all drugs (balanced accuracy values ≥0.75). The GTMs represent data in the form of illustrative 2D maps of the genomic space and allow for antibiotic-wise comparison of resistance phenotypes. In addition to that, the maps were found to be useful for the analysis of genetic determinants responsible for drug resistance based on the data from the PATRIC database. Overall, the GTM-based methodology is a useful tool for the illustrative exploration of the genomic sequence space and modelling AMR and can be used as a tool complementary to the existing ML methods for AMR prediction.Availabilityhttps://doi.org/10.5281/zenodo.7101559