Prediction of Antimicrobial Resistance in Gram-Negative Bacteria From Whole-Genome Sequencing Data

• Main Authors: Pieter-Jan Van Camp, David B. Haslam, Aleksey Porollo
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2020-05-01
• Series: Frontiers in Microbiology
• Subjects: antimicrobial resistance; antibiotic resistance; whole-genome sequencing; machine learning; prediction; genotype-phenotype relationship
• Online Access: https://www.frontiersin.org/article/10.3389/fmicb.2020.01013/full

BackgroundEarly detection of antimicrobial resistance in pathogens and prescription of more effective antibiotics is a fast-emerging need in clinical practice. High-throughput sequencing technology, such as whole genome sequencing (WGS), may have the capacity to rapidly guide the clinical decision-making process. The prediction of antimicrobial resistance in Gram-negative bacteria, often the cause of serious systemic infections, is more challenging as genotype-to-phenotype (drug resistance) relationship is more complex than for most Gram-positive organisms.Methods and FindingsWe have used NCBI BioSample database to train and cross-validate eight XGBoost-based machine learning models to predict drug resistance to cefepime, cefotaxime, ceftriaxone, ciprofloxacin, gentamicin, levofloxacin, meropenem, and tobramycin tested in Acinetobacter baumannii, Escherichia coli, Enterobacter cloacae, Klebsiella aerogenes, and Klebsiella pneumoniae. The input is the WGS data in terms of the coverage of known antibiotic resistance genes by shotgun sequencing reads. Models demonstrate high performance and robustness to class imbalanced datasets.ConclusionWhole genome sequencing enables the prediction of antimicrobial resistance in Gram-negative bacteria. We present a tool that provides an in silico antibiogram for eight drugs. Predictions are accompanied with a reliability index that may further facilitate the decision making process. The demo version of the tool with pre-processed samples is available at https://vancampn.shinyapps.io/wgs2amr/. The stand-alone version of the predictor is available at https://github.com/pieterjanvc/wgs2amr/.