AI-CURA, an automated LLM workflow for high-accuracy genetic variant classification - PubMed
5 hours ago
- #genetic variant classification
- #clinical genetics
- #large language models
- AI-CURA is an automated LLM workflow designed for high-accuracy genetic variant classification, nearly fully automating the process according to ACMG/AMP guidelines and ClinGen recommendations.
- It integrates automated evidence assessment for non-literature-based criteria with a separate LLM-supported assessment of literature-based evidence, enhancing clinical workflow designs in genetics.
- Two advanced LLMs, DeepSeek-R1 and o3-mini-high, were tested, with DeepSeek-R1 outperforming in summarizing literature-derived evidence, achieving high sensitivity and 100% specificity in interpreting ACMG rules.
- Testing with 150 variants curated by ClinGen experts showed high concordance between DeepSeek-R1 and human curators in final diagnosis.
- AI-CURA also demonstrated utility in classification reanalysis using 150 ClinVar variants with conflicting interpretations, offering a framework for automated variant classification and reanalysis in genetic disease diagnosis.