CABBAGE: Comprehensive Assessment of Bacterial-Based AMR prediction from Genotypes

Antimicrobial resistance (AMR) is a growing public health threat. Not only does it create the risk of untreatable bacterial infections, but also increases the likelihood of serious complications from the essential procedures of modern medicine, such as transplants, chemotherapy, or even childbirth. Our understanding of AMR has improved dramatically with the advent of next-generation sequencing (NGS), enabling us to gain insights into the mechanisms of AMR and identify these mechanisms directly from clinical samples, such as blood, sputum, or urine. This subsequently allows us to target each infection with the precise treatment specific to it and improve the efficacy of the existing antimicrobials, as almost no new classes of antimicrobials have entered the market in the last 40 years. However, the insights necessary for understanding the mechanisms causing AMR and identifying them directly from samples are highly dependent on the data used for their analysis. Efforts to carry out large-scale analyses have been hindered by multiple factors; one key factor is that publicly available combined datasets containing NGS data and AMR data are presented in a variety of formats, and follow different conventions with regards to their interpretation. This project, Comprehensive Assessment of Bacterial-Based Antimicrobial resistance prediction from GEnotypes (CABBAGE), attempts to provide a solution by collecting and curating all the publicly available data containing both NGS information and AMR information, transforming it into a standard format, and making it accessible it to the research community and the general public as a resource to be maintained by the European Bioinformatics Institute (EBI). In a pilot project, we collected more than three times as much data as is currently available from the single largest public database, in a reconciled, uniform format. We have also conducted an initial benchmarking for one of the pathogens, Klebsiella pneumoniae, showing both the importance and the subtleties of interpretable prediction of AMR phenotypes. Whilst this initial effort required considerable manual processing, one of the project's goals will be to automate the process so that the results of future studies can be directly incorporated into this database, and work with stakeholders such as the World Health Organisation to facilitate the adoption of our standards. Additional goals of the project include investigating the limitations of the ability of NGS information to predict AMR, to systematically compare existing approaches for carrying out these predictions, to set out quality criteria for the evidence required to identify novel resistance mechanisms in a data-driven way, to check experimentally that any mechanisms discovered according to these criteria are accurate, and lastly, to create initial draft catalogues of genomic determinants of AMR for each of the 7 most commonly found bacterial pathogens on the WHO's priority pathogens list. The outcomes from this project will include a comprehensive resource on NGS and AMR, a better understanding of the limits restrictions of AMR prediction predictive ability from NGS information, and quality criteria required for the evidence supporting novel resistance mechanisms. These outcomes will serve several communities working on AMR. First, they will help diagnostics developers assess the performance of their proposed tests. Second, they will help public health microbiologists assess the risk posed by specific infections in real-time. Lastly, they will help clinicians prescribe the optimal drug regimen for each infection they treat.