Working project title
Evaluating and Improving the Specificity of Stool Microbiome AI for Colorectal Cancer in the Presence of Crohn’s Disease
Status Proposal and protocol development
Research question
Do models developed to distinguish colorectal cancer from control samples incorrectly flag Crohn’s disease samples, and can incorporating inflammatory-disease information improve specificity while preserving cancer sensitivity?
Why ask it
A model trained only on colorectal cancer and control samples has not been asked to tell cancer apart from other conditions that affect the gut. Inflammatory bowel disease, including Crohn’s disease, is also associated with changes in the gut microbiome. This project sets out to measure whether that leads to false alarms, and whether adding inflammatory-disease information helps without causing more cancers to be missed.
Where the project stands
Only the items marked Done or In progress have started. No experiments, dataset downloads, validation, or publications have happened.
- Done Define the research question
- In progress Develop the proposal and protocol
- Planned Harmonize cohorts and disease labels
- Planned Check patient identities and repeated samples
- Planned Build reproducible ETL pipelines
- Planned Compare cancer-versus-control models with approaches incorporating Crohn’s disease information
- Planned Evaluate on held-out studies
- Planned Report false positives, missed cancers, calibration, and limitations
- Planned Version code and document experiments throughout
Candidate data resource
The candidate resource is the public processed stool microbiome profiles and associated metadata in curatedMetagenomicData. It has not been selected yet: dataset suitability and patient-level labels still need to be audited.