Are there changes in the virus during the attenuation process that can be used to improve future vaccines?

Using deep learning (DL) to identify patterns of attenuation in existing deep sequencing datasets

This work will utilize datasets generated from a previous BBSRC-funded project (BB/L003988/1) investigating the attenuation of 2 IBV viruses in parallel (a commercial vaccine and a lab adapted strain). We have generated 52 ultra-deep whole IBV genome sequence datasets (average coverage depth: 1,000-10,000 reads) from different stages of the attenuation process. In these datasets, variants have been identified both exclusively in attenuated viruses, and shared between pathogenic and attenuated viruses. High coverage datasets will help eliminate defective interfering viruses from the analysis. We will use a recently developed supervised deep learning tool to identify patterns in both variants already identified and in sequence data from each sequenced passage. This will identify genomic signatures or patterns present in either or both pathogenic and attenuated viruses. Similar work with DL and viruses has been published previously. We will use pre-existing phenotypic data, including viral load (quantitative RT-PCR), viral replication kinetics and ciliostasis tests, to define labels for training. Using bioinformatics techniques, we will also characterise regions of secondary, tertiary and protein coding structures and recombination hotspots that may play roles in linking known SNPs in both pathogenic and attenuated groups, comparing with field data from different pathogenic IBV viruses.

Anticipated outcomes

WP1 will use deep learning to identify patterns/genomic signatures present in viruses with attenuated or pathogenic phenotypes or shared across both. We will correlate these predictions with bioinformatics analysis characterising location relative to secondary and tertiary structure and genetic distance from known pathogenic IBV isolates. Identified markers will be analysed further in WP4.