P024

1:University of Oxford; 2:University of Bath

Huge progress towards translating whole genome sequencing (WGS) into tuberculosis diagnostics and surveillance has been made in the last few years with the advent of the WHO catalogues of resistance-associated variants (RAVs) for tuberculosis. The process to build the catalogue is, unfortunately, slow and not reproducible. Here we describe processing 53,700 clinical tuberculosis samples with WGS data and drug susceptibility testing (DST) results that are publicly available via the CRyPTIC project and automatically generating a catalogue with minimal user intervention. FASTQ files were processed by an online cloud processing platform. Resulting data tables were ingested by catomatic, a freely available Python package, producing a catalogue for 15 drugs. We curated an independent Validation dataset, allowing us to truly assess the performance of the resulting catalogue. The whole process took less than six weeks from FASTQ to sensitivity/specificity graphs. Our catalogue performed similarly to both editions of the WHO catalogue for all drugs without adding any manually-defined expert rules. Since catalogue building is automatic, we were able to vary the proportion of reads needed to identify an RAV and this improved performance, with some drugs (e.g. amikacin) affected more than others. As more M. tuberculosis samples are amassed it is becoming possible to generate resistance catalogues rapidly and automatically which will permit us to both respond more quickly and potentially tailor catalogues for specific geographies. Approaches like this could also be applied to common NTM species as WGS/pDST datasets for these become available.