Antimicrobial Resistance Prediction in PATRIC and RAST

• Published in: Scientific reports Vol. 6; no. 1; p. 27930
• Main Authors: Davis, James J., Boisvert, Sébastien, Brettin, Thomas, Kenyon, Ronald W., Mao, Chunhong, Olson, Robert, Overbeek, Ross, Santerre, John, Shukla, Maulik, Wattam, Alice R., Will, Rebecca, Xia, Fangfang, Stevens, Rick
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 14.06.2016; Nature Publishing Group
• Subjects: 631/114; 631/114/129/2043; 631/114/1305; adaptive boosting; Algorithms; Anti-Bacterial Agents - therapeutic use; Antibiotics; Antimicrobial agents; Bacterial Infections - drug therapy; BASIC BIOLOGICAL SCIENCES; Clinical Decision-Making; Computational Biology; computational biology and bioinformatics; Data Curation; Databases, Genetic; Drug resistance; Drug Resistance, Microbial - genetics; Genes; genetic databases; genome annotation; Genome, Bacterial - genetics; Genomes; Genomics; Genotype & phenotype; Humanities and Social Sciences; Humans; Machine Learning; Medical research; Metadata; Microbial Sensitivity Tests; Molecular Sequence Annotation; multidisciplinary; National Institutes of Health (U.S.); Pathogens; Prognosis; random forest; Science; Streptococcus infections; support vector machines; United States; United States > US
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/srep27930

The emergence and spread of antimicrobial resistance (AMR) mechanisms in bacterial pathogens, coupled with the dwindling number of effective antibiotics, has created a global health crisis. Being able to identify the genetic mechanisms of AMR and predict the resistance phenotypes of bacterial pathogens prior to culturing could inform clinical decision-making and improve reaction time. At PATRIC ( http://patricbrc.org/ ), we have been collecting bacterial genomes with AMR metadata for several years. In order to advance phenotype prediction and the identification of genomic regions relating to AMR, we have updated the PATRIC FTP server to enable access to genomes that are binned by their AMR phenotypes, as well as metadata including minimum inhibitory concentrations. Using this infrastructure, we custom built AdaBoost (adaptive boosting) machine learning classifiers for identifying carbapenem resistance in Acinetobacter baumannii , methicillin resistance in Staphylococcus aureus , and beta-lactam and co-trimoxazole resistance in Streptococcus pneumoniae with accuracies ranging from 88–99%. We also did this for isoniazid, kanamycin, ofloxacin, rifampicin, and streptomycin resistance in Mycobacterium tuberculosis , achieving accuracies ranging from 71–88%. This set of classifiers has been used to provide an initial framework for species-specific AMR phenotype and genomic feature prediction in the RAST and PATRIC annotation services.