Comprehensive Antibiotic Resistance Profiling

A total of 427,495 bacterial genomes consisting of 47,582,748 sequences (NCBI GenBank database, 2019-10-22) [32 ] were analyzed using fARGene (v0.1, default parameters). We used fARGene in this study since it has been shown to have a high performance and its predictions have been experimentally verified on several occasions [12 , 13 (link), 15 , 16 (link), 33 , 34 (link)]. fARGene was executed using 17 hidden Markov model gene profiles for ARGs conferring resistance to five major classes of antibiotics: for $\beta$ -lactams, we defined gene classes A, B1/B2, B3, and D [13 (link), 33 ]; for aminoglycosides, gene classes $aac(2^{\prime })$ , aac(3), $aac(6^{\prime })$ , $aph(2^{″})$ , $aph(3^{\prime })$ , and aph(6); for macrolides, gene classes erm and mph [12 ]; for quinolones, gene class qnr [16 (link)]; and for tetracyclines, gene classes efflux pumps, inactivating enzymes (monooxygenases), and ribosomal protection genes (RPGs) [15 ] (downloaded from https://github.com/fannyhb/fargene). All matches satisfying the previously reported model-specific significance thresholds for full-length genes were considered to be putative ARGs and stored for further analysis [12 , 13 (link), 16 (link), 33 , 35 (link)].