A Semantic Model of Genetic Evidence: A Step Toward Bridging the Basic-Science-Clinic Gap

The paper addresses a critical gap in how genetic evidence is currently documented: existing standards like FHIR Evidence and GA4GH focus on clinical trials or single genetic variants, but don't capture the detailed structure of evidence from basic research—the foundational studies that inform clinical decisions about genetic variants.

The researchers developed a new semantic model (a standardized way of organizing information) specifically designed for genetic evidence that can be validated using automated rules (SHACL schema), which is important for ensuring AI systems interpret genetic research consistently and reliably.

In a pilot study analyzing six genetics papers, the team tested their model with both human curators and AI systems, deliberately keeping human-authored annotations separate from AI-drafted ones—a responsible practice that maintains transparency about which interpretations came from human experts versus automated systems.

The model produced 28 evidence items and 95 assertions tied to specific sources in the papers, demonstrating that detailed, traceable documentation of genetic evidence is feasible and could improve how clinical decisions about genetic variants are made and justified.

The authors frame this work as infrastructure for "trustworthy, AI-ready" variant interpretation, suggesting that standardized, validated ways of representing evidence are necessary before AI systems can reliably assist in translating basic genetic research into clinical recommendations.