Antimicrobial resistance (AMR) is a growing global health threat. In 2019, about 1.27 million deaths were linked to bacteria that no longer respond to antibiotics and resistance has been increasing in Canada. The standard test to determine whether bacteria are resistant is slow because it relies on growing the bacteria in the lab. It also can’t easily explain why the bacteria are resistant. Faster and more informative tools are urgently needed so that healthcare providers can choose the right antibiotics sooner.  

New technologies such as whole genome sequencing (WGS) and metagenomic sequencing can quickly read bacterial DNA and may replace older tests. Next generation sequencing, such as nanopore sequencing, has made this faster and more practical for hospitals. However, even with DNA data, predicting resistance remains challenging because genes don’t always determine how bacteria behave in real world conditions.. 

BugSeq is a cloud-based platform already used by public health labs in Canada and the U.S. Its AMR prediction system, BugAMR, uses machine learning and a manually curated database to predict antibiotic resistance more accurately than many existing tools. Despite this progress, gaps remain — especially for certain bacteria species and antibiotic classes.  

In this project, BugSeq aims to improve prediction accuracy by adding new features such as gene copy number (how many copies of a resistance gene a bacterium has) and by discovering new resistance markers through large-scale genetic analysis. They will also use large language models to rapidly scan scientific literature for additional resistance genes missing from public databases. 

To achieve this, BugSeq will train its system using a set of over 58,000 high-quality bacterial genomes. The improved system will then be validated through computational testing and real-world clinical testing at Providence Health Care, Vancouver Coastal Health and Johns Hopkins University.