RareSim
Similarity-based AI for Rare Diseases

From clinical notes to a rare disease diagnosis.

Rare diseases take an average of five years to diagnose. RareSim reads the clinical notes that already exist, matches what it finds against more than 6,000 known conditions, and keeps clinicians in the loop.

Clinical notesfree text, as writtenExtractphenotypes in contextMatchdiseases and past patientsCandidatesranked ORPHAcodes
Approach

Extract, match, improve

  1. Extract

    We read clinical notes and pull out the patient’s phenotypes — the observable signs and symptoms — keeping track of context like negation, family history and timing.

  2. Match

    We compare that phenotype profile against known rare diseases and against patients who already have a diagnosis, and return a ranked list of candidates.

  3. Improve

    Clinicians correct what we get wrong. That feedback flows back into both earlier steps, so the system gets sharper the more it is used.

Publications

Research output

The code accompanying our papers is available on GitHub.

Our first papers are in preparation and will be listed here with DOIs and author copies.

News

Updates

  • RareSim launches its project websiteMilestone

Rare questions welcome

About the method, the papers, or the code.

[email protected]