Agentic discovery of blood biomarker from distilled private health records
Researchers release a privacy-preserving scorer so agents can find CBC biomarkers without patient data.
A paper describes distilling evidence from Clalit Health Services records covering more than 5.4 million patients into a released scoring tool. For each of 13 immune-mediated diseases, a graph attention network was trained inside the data boundary to predict the case-control AUC of candidate complete-blood-count expressions, and only the weights were released. In external validation, agent-discovered expressions improved on literature-seeded starting points by a median of 4.18 AUC percentage points. Reranking candidates from three frontier research tools improved on their first choices in most comparisons across three independent cohorts.
- Scorer trained on Clalit records from over 5.4 million patients; only weights released.
- Graph attention networks predict AUC for CBC expressions across 13 diseases.
- Agents beat literature seeds by a median of 4.18 AUC points.
- Reranking three tools’ candidates helped in most cohort comparisons.
- Organizations
- Clalit Health Services
From the original article
1 min read at huggingface.co
Routine complete blood counts (CBCs) could yield new biomarkers, but the private records needed to evaluate candidates cannot be shared with frontier language model agents that excel at discovery. We distilled the evidence held in the Clalit Health Services panel of over 5.
By Seffi Cohen, Liat Antwarg Friedman, Amir Anisman · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.