Predicting Antimicrobial Resistance Using Conserved Genes

Abstract

Abbreviations AIartificial intelligence CI confidence interval ME major error, susceptible genomes predicted to be resistant MIC minimum inhibitory concentration PATRIC Pathosystems resource integration center PLF PATRIC local protein family RAST Rapid annotation using subsystem technology RF Random Forest SR susceptible and resistant VME very major error, resistant genomes predicted to be susceptible XGB XGBoost Abstract A growing number of studies have shown that machine learning algorithms can be used to accurately predict antimicrobial resistance (AMR) phenotypes from bacterial sequence data. In these studies, models are typically trained using input features derived from comprehensive sets of known AMR genes or whole genome sequences. However, it can be difficult to determine whether genomes and their corresponding sets of AMR genes are complete when sequencing contaminated or metagenomic samples. In this study, we explore the possibility of using incomplete genome sequence data to predict AMR phenotypes. Machine learning models were built from randomly-selected sets of core genes that are held in common among the members of a species, and the AMR-conferring genes were removed based on their protein annotations. For Klebsiella pneumoniae, Mycobacterium tuberculosis, Salmonella enterica, and Staphylococcus aureus, we report that it is possible to classify susceptible and resistant phenotypes with average F1 scores ranging from 0.80-0.89 with as few as 100 conserved non-AMR genes, with very major error rates ranging from 0.11-0.23 and major error rates ranging from 0.10-0.20. Models built from core genes have predictive power in the cases where the primary AMR mechanism results from SNPs or horizontal gene transfer. By randomly sampling non-overlapping sets of core genes for use in these models, we show that F1 scores and error rates are stable and have little variance between replicates. Potential biases from strainspecific SNPs, phylogenetic sampling, and imbalances in the phylogenetic distribution of susceptible and resistant strains do not appear to have an impact on this result. Although these small core gene models have lower accuracies and higher error rates than models built from the corresponding assembled genomes, the results suggest that sufficient variation exists in the core non-AMR genes of a species for predicting AMR phenotypes. Overall this study suggests that building models from conserved genes may be a potentially useful strategy for predicting AMR phenotypes when genomes are incomplete.

A Genome-Based Model to Predict the Virulence of Pseudomonas aeruginosa Isolates

“…Together, these findings provide additional evidence that a broad genetic fingerprint, rather than individual virulence or antivirulence factors, is being used to classify strains as having high or low virulence. Furthermore, it is consistent with a recent finding that antimicrobial resistance in several species can be accurately predicted by only considering variation in a small subset of core genes (and excluding known resistance genes) (35).…”

Predicting Antimicrobial Resistance Using Partial Genome Alignments

“…When the Salmonella models are recomputed using the chromosome of Salmonella enterica serovar Typhi CT18 as the reference, we also observe nearly identical results (see Table S1D in the supplemental material), indicating that the choice of the reference chromosomes has little impact. Overall, these AUCs and corresponding model statistics, including F1 scores and error rates (see Tables S1E to S1G in the supplemental material), are consistent with previous studies that have used k-mers and AMR genes as input (14,15,26).…”

“…Since the machine learning models learn from nucleotide similarity that has been observed across samples in the training set, some of the high accuracies are likely due to similar sequences occurring in both the training and testing sets, which may obscure the more broadly conserved nucleotide signatures relating to, or correlating with, resistance. This has been observed previously, and several studies have tried to balance strains based on phylogeny to reduce this effect in the ML models (10,11,21,26). In essence, having related genomes in the training and testing sets can improve accuracies, but may also potentially reduce the generalizability of the models if the overall phylogenetic distribution of the training set is biased.…”

“…In previous work, Nguyen et al demonstrated that AMR phenotypes could be predicted using sets of conserved genes that are held in common among the members of a species, even if they are not known to be involved in AMR (26). The reasons relatively high accuracies are observed in the non-AMR genes are not entirely clear.…”

Shotgun-Metagenomics Based Prediction Of Antibiotic Resistance And Virulence Determinants In Staphylococcus Aureus From Prosthetic Joint Tissue On Blood Culture Bottles

“…Strategies for predicting AMR phenotypes in polymicrobial samples present an interesting challenge 36 . We tested the binning approach for the prediction of ARGs in S. aureus and we found that it gives similar results as prediction of ARGs in monomicrobial samples for most of the genes with exception of the blaZ gene.…”