Data Challenge QUB team LD EC

Title: Surveilling the global AMR crisis by predicting AMR phenotype from the MERCK SMART-surveillance dataset.
Antimicrobial resistance (AMR) poses a significant global health, with projections estimating up to 10 million AMR-related deaths annually by 2050. Gaining a deeper understanding of how genomic mechanisms form AMR phenotypes is essential to curbing the spread of AMR. While AMR is often attributed to the presence of a singular resistance gene, our previous work has demonstrated that despite this, AMR phenotypes arise from complex interactions between multiple genes, including both defined AMR genes and accessory genes not known to be involved in resistance.
If we can gain a comprehensive understanding of how the entire genome relates to AMR phenotypes, this will shed light on how AMR mechanisms emerge. Using decision tree modelling, we have previously shown that the routes to resistance are not only complex involving multiple genes, but these routes can be species-specific.
Using Merck's SMART Surveillance program dataset, we aim to expand these models to predict AMR phenotypes for a wider range of antibiotics and bacterial taxonomy. We can use the available genomic information to provide features for our machine-learning models. The role of mobile genetic elements (e.g. plasmids and prophages) may also play a key role in AMR proliferation; we can predict these regions and investigate the interactions between chromosomal and extra-chromosomal genes. This can establish how known AMR genes and accessory genes influence AMR phenotypes.
Our use of interpretable machine-learning models enables the identification of genes not typically associated with AMR phenotypes, offering novel targets for drug discovery. Elucidating resistance mechanisms at this large scale will help to address AMR global transmission and present potential avenues to help mitigate the spread.