An agentic system for rare disease diagnosis with traceable reasoning

N/A (multi-gene AI system)

Multi-step agentic architecture with traceable reasoning: AI integrates HPO phenotypic data, WES/WGS candidate variants, OMIM/ClinVar/HGMD knowledge, and produces a prioritized diagnosis with step-by-step audit trail; validated on undiagnosed rare disease cohorts

Presentation of an agentic AI system for rare disease diagnosis, based on traceable multi-step reasoning. The system integrates patient HPO phenotypic data, candidate variants from WES/WGS, and genomic databases (OMIM, ClinVar, HGMD, HPO). It produces a prioritized diagnosis with a human-readable step-by-step audit trail, enabling transparent medical supervision. Validated on diagnostic odyssey patients, it demonstrates superior performance over previous non-audit AI systems while meeting EU AI Act explainability requirements.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

First diagnostic AI system for rare diseases with truly traceable and auditable reasoning, published in Nature. The agentic aspect (sequential decision-making with contextual memory) represents a qualitative leap over static prioritization models. Implications for rare genetics laboratories are major: integrable into unresolved case review workflows. Medical supervision remains essential.

Nature (top journal) +3; agentic system with explainable reasoning (XAI) +2; validated on real undiagnosed cohorts +2; impact on diagnostic odyssey +2