GAMBIT (Genomic Approximation Method for Bacterial Identification and Tracking): A methodology to rapidly leverage whole genome sequencing of bacterial isolates for clinical identification

Abstract

AbstractWhole genome sequencing of clinical bacterial isolates has the potential to transform the fields of medicine and public health, with particular impact on molecular epidemiology, infection control and assessing the spread of antibiotic resistance. To realize this potential, bioinformatic software needs to be developed that meets the quality standards of a diagnostic test to allow the reporting of identification results. Our research group has developed a methodology (GAMBIT: Genomic Approximation Method for Bacterial Identification and Tracking) using k-mer based strategies for identification of bacteria based on whole genome sequence reads. GAMBIT incorporates this algorithm with a highly curated searchable database of genomes, which is a subset of the NCBI RefSeq assembly database. In this manuscript, we describe the validation of the scoring methodology, robustness to chosen parameters, establishment of confidence thresholds and the curation of the reference database. We discuss a validation set with GAMBIT deployed as a laboratory-developed test at the Alameda County Public Health Laboratory in Oakland, California. Three advancements were required to build upon existing k-mer based strategies to allow GAMBIT to possess the quality control parameters desired for its use as a diagnostic laboratory-developed test. Firstly, we innovated the data structure used to store the database of known bacterial genomes. This allowed us to store 48,224 bacterial genomes in a k-mer based database—a majority of the bacterial genomes in the NCBI RefSeq database at the time of development. Secondly, curation of the NCBI RefSeq database was required to remove ambiguous or potentially incorrectly labeled bacterial genomes to greatly increase confidence in positive matches. Lastly, we used this curated version of the NCBI RefSeq database and our scoring method to generate confidence thresholds for identification. Thus, the end-user does not rely simply on the closest match, but is informed whether that closest match exceeds a threshold for highly confident identification. This method greatly reduces or eliminates false identifications which are often detrimental in a clinical setting.