Blood biomarker discovery tool improves disease detection accuracy
Agentic discovery of blood biomarker from distilled private health records
Artificial Intelligence
Summary
Finding new signs of diseases in routine blood tests can help doctors diagnose illnesses earlier, but patient privacy makes sharing detailed medical records hard. The authors created a tool that learns from millions of private health records without exposing any personal data. This tool helps smart computer agents suggest and score new blood test markers for immune diseases more accurately than previous methods. It works by sharing only learned patterns, not the original sensitive information.
What this means in practice
- •For clinical data teams: Use a privacy-preserving scoring tool to evaluate candidate blood biomarkers for immune diseases without sharing raw patient data.
- •For medical device developers: Improve blood test diagnostic algorithms by integrating agent-discovered biomarkers identified through a privacy-safe external validation method.$Commercial implications: Enables development of more accurate diagnostic software for blood analyzers without violating patient confidentiality.
Authors
Seffi Cohen, Liat Antwarg Friedman, Amir Anisman, Ruth Johnson, Michelle M. Li, Ayush Noori, Ben Reis, Ran Balicer, Noa Dagan, Marinka Zitnik
Abstract
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 5.4 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 CBC expressions, and only the trained weights were released. The tool grounds an agent's propose-score-refine loop in real-world data without exposing any patient data. In external validation, agent-discovered expressions improved on their literature-seeded starting points by a median of 4.18 AUC percentage points, and across three independent cohorts, reranking the candidates of three frontier research tools improved on their first choices in most comparisons, with gains that varied by cohort. The released scorer supports privacy-preserving biomarker hypothesis generation.