VAMPr: VAriant Mapping and Prediction of antibiotic resistance via explainable features and machine learning

Abstract

Antimicrobial resistance (AMR) is an increasing threat to public health. Current methods of determining AMR rely on inefficient phenotypic approaches, and there remains incomplete understanding of AMR mechanisms for many pathogen-antimicrobial combinations. Given the rapid, ongoing increase in availability of high density genomic data for a diverse array of bacteria, development of algorithms that could utilize genomic information to predict phenotype could both be useful clinically and assist with discovery of heretofore unrecognized AMR pathways. To facilitate understanding of the connections between DNA variation and phenotypic AMR, we developed a new bioinformatics tool, variant mapping and prediction of antibiotic resistance (VAMPr), to (1) derive gene ortholog-based sequence features for variants; (2) interrogate these explainable gene-level variants for their known or novel associations with AMR; and (3) build accurate models to predict AMR based on whole genome sequencing data. Following the Clinical & Laboratory Standards Institute (CLSI) guidelines, we curated the publicly available sequencing data for 3,393 bacterial isolates from 9 species along with AMR phenotypes for 29 antibiotics. We detected 14,615 variant genotypes and built 93 association and prediction models. The association models confirmed known genetic antibiotic resistance mechanisms, such as blaKPC and carbapenem resistance consistent with the accurate nature of our approach. The prediction models achieved high accuracies (mean accuracy of 91.1% for all antibiotic-pathogen combinations) internally through nested cross validation and were also validated using external clinical datasets. The VAMPr variant detection method, association and prediction models will be valuable tools for AMR research for basic scientists with potential for clinical applicability.

Machine learning in the clinical microbiology laboratory: has the time come for routine practice?

“…The method gave good performances (94e99% correct predictions) in E. coli, K. pneumoniae and Acinetobacter baumannii. The authors provided a proof-of-concept of the test directly from positive blood cultures with a turn-around time of <4 h. Other ML systems used WGS to combine the identification and antibiotic susceptibility testing of microorganisms: Drouin et al reported an ML system that can identify 12 species and susceptibility to 56 antibiotics [61], Kim et al nine species and 29 antibiotics [62].…”

“…Moreover, the systems still required the most time-consuming step in the diagnosis which is the technical manipulation of the stools (fresh state, concentrations, preparations, colourations). In bacteriology, most ML systems targeted either bacterial identification or antibiotic susceptibility testing for a limited number of species (median 3, IQR 7e15), but optimal tools should be able to perform different tasks on a broad scope of species using the same data, as has been done in some recent articles [61,62].…”

Recent evolutions of machine learning applications in clinical laboratory medicine

“…The lengthy nature of conventional methods for bacterial identification and susceptibility testing purposes has resulted in the empirical use of broad-spectrum antibiotics and led to the spread of resistance [85]. Aiming at rapid and reliable detection of AMR, numerous studies have recently been reported on the use of ML algorithms in combination with the output of several analytical techniques, such as MALDI-TOF MS [89,96,[110][111][112][113][114][115][116][117][118], vibrational spectroscopy [102,103,119], whole-genome sequencing [120][121][122][123][124], a microscopy-based platform [125], and acousticenhanced flow cytometry [126].…”

“…The authors showed that the predictive accuracy of the applied tree models was the lowest for resistance to third-line drugs such as para-aminosalicylic acid and Dcycloserine (AUROC below 0.85) and highest for resistance to first-line drugs such as amikacin, ciprofloxacin, kanamycin, moxifloxacin, and multi-drug resistant tuberculosis (AUROC above 0.96). Furthermore, a novel bioinformatic tool for variant mapping and prediction of antibiotic resistance (VAMPr) has been reported by Kim et al [124] with a mean accuracy of 91.1% for all antibiotic-pathogen combinations.…”

Protocol for detection of bacterial proteins involved in efflux mediated antibiotic resistance (ARE) and their sub-families

“…The e ux proteins are responsible for multi-drug resistance in many microbial pathogens [1][2][3][4][5] . In the past several highly accessed and useful antibiotic resistance databases have been established to catalogue the known antibiotic resistance genes at both the whole genome as well as at genes/proteins levels [6][7][8][9][10][11][12][13][14][15] . But an in-silico tool to predict and annotate e ux proteins responsible for antibiotic resistance (ARE) has not been developed yet.…”