Machine-Learning Model for Prediction of Cefepime Susceptibility in Escherichia coli from Whole-Genome Sequencing Data

Abstract

The declining cost of performing bacterial whole-genome sequencing (WGS) coupled with the availability of large libraries of sequence data for well-characterized isolates have enabled the application of machine-learning (ML) methods to the development of nonlinear sequence-based predictive models. We tested the ML-based model developed by Next Gen Diagnostics for prediction of cefepime phenotypic susceptibility results in Escherichia coli .

Multicenter Evaluation of the BIOFIRE Blood Culture Identification 2 Panel for Detection of Bacteria, Yeasts, and Antimicrobial Resistance Genes in Positive Blood Culture Samples

“…When the presence or absence of a single gene marker (e.g., CTX-M) is used to guide empirical antimicrobial therapy, it is important to consider not only the sensitivity of an assay but also its negative predictive value, which is dependent upon local prevalence of overall resistance and also resistance mechanisms ( 11 ). In the future, characterization and analysis of full genomic data rather than single gene markers could be used to more accurately predict phenotypic susceptibility and resistance ( 12 ). With the addition of the expanded Enterobacterales AMR targets, local laboratories in collaboration with local antimicrobial stewardship programs will be able to design optimized empirical antibiotic regimens that can be informed by the BIOFIRE BCID2 Panel AMR results.…”

Status check: next-generation sequencing for infectious-disease diagnostics

“…Economic analyses have demonstrated the value of prospective WGS over traditional reactive approaches to identify and contain hospital-acquired infection clusters ( 1 , 2 ). Furthermore, data from the whole genome and/or resistome (all AMR genes) of a bacterial pathogen combined with machine-learning approaches have enabled predictions of the phenotypic susceptibility profile — with similar accuracy as traditional growth-based approaches ( 3 ). Applying rapid WGS could provide more timely results to guide patient management ( 4 ).…”

“…It is important to note that there is poor negative predictive value, as the lack of detection of AMR does not rule out the presence of an AMR gene. Also, for polymicrobial specimens, association of the AMR gene with a particular pathogen is difficult as it can be impossible to link the AMR marker to a specific organism, unless long read sequencing provides genetic context to the AMR gene ( 3 ). Furthermore, the AMR gene cannot be linked to one pathogen over another if multiple pathogens known to harbor the AMR gene are detected, or for cases in which benign microbiota harbor AMR genes, which is a scenario that can lead to predictions of false resistance.…”

Analytical and Clinical Validation of Direct Detection of Antimicrobial Resistance Markers by Plasma Microbial Cell-free DNA Sequencing

“…WGS can be used either to detect AMR in culture isolates or directly from clinical specimens. However, WGS is predicated on a detailed understanding of the relationship between genotype and resistance phenotype that is presently incomplete and may require advanced machine learning algorithms to improve the ability to accurately predict AMR from WGS data (6, 7). Resistance mechanisms relating to the presence or absence of specific genes or well-characterized target site mutations will be more straightforward to interpret than those relating to alterations in gene expression.…”