UC San Diego Undergraduate Library Research Prize

First Place Winner 2023 Life Sciences, Physical Sciences and Engineering Category, nominated by Weg M. Ongkeko (Department of Surgery).

Inflammatory bowel diseases (IBD) are disorders of the gastrointestinal system affecting a considerable portion of Western populations. With nearly 6.8 million individuals affected globally, the ability to accurately and efficiently diagnose the disease in patients is crucial. The gastrointestinal microbiome composition has been associated with several diseases, disorders, and cancers. Several microbial signature-based machine learning IBD predictive models have been developed but have yet to be implemented clinically due to poor external accuracy. This study analyzes the microbial compositions of 1612 stool samples from 665 patients across seven distinct shotgun metagenomic cohorts in constructing an IBD diagnostic model capable of high predictive accuracy and robustness. We created a 14-species diagnostic ML panel and validated the performance of the model on 3 independent patient cohorts. Testing the ML model against other microbiome-associated diseases, such as colorectal cancer and type 2 diabetes, suggested that the diagnostic model is specific to IBD. The model's reliance on abundance levels of only 14 biomarker species, in addition to its non-invasive nature, makes diagnosis relatively inexpensive and convenient. Furthermore, the signature is stable across different patient cohorts, sequencing pipelines, and sample preparation methods. To examine potential mechanisms by which the microbiome influences IBD, we identified 23 metabolic pathways that were consistently differentially enriched across cohorts, 2 microbes of the greatest contribution to differences in pathway abundance, and metabolites correlated to the presence of the 14 biomarker species.