P113

Whole genome sequencing is faster, cheaper and, for many drugs, as accurate as traditional phenotypic testing for tuberculosis and is increasingly being adopted by Public Health Agencies and hospitals. This shift is being encouraged by the WHO – they have recently released the second edition of their catalogue of resistance associated genetic variants, yet it remains difficult in practice to apply the information contained within the WHO catalogue. One consequence is it is challenging for pipeline and tool developers to assure that their implementation of the WHO catalogue is faithful to the original.

Here we calculate the performance for all the drugs in the WHO catalogue using two datasets; one derived from the CRyPTIC project (described in another abstract) and a second complementary dataset that “fills in” the missing compounds. Our results are comparable to those reported by the WHO on their training set, which is pleasing if a little surprising given the datasets are different. We then analyse the discrepancies between the genotype and phenotype in our dataset but are unable to explore the discrepancies with the reported WHO figures since their dataset is not publicly available. Finally, we describe the key design principles, such as ensuring no identifiable information in the cloud, and the architectural designs that underpin GPAS, for example the use of Kubernetes to grow the number of samples that can be processed concurrently, as demand allows.