AI-CURA, an automated LLM workflow for high-accuracy genetic variant classification

Large language models (LLMs) have been extensively tested for incorporation into medical applications in recent years; however, their potential in clinical genetics, particularly in diagnosing rare diseases, remains underexplored. Recent advancements in LLMs have improved their reasoning capabilities and transparency, facilitating enhancements in clinical workflow designs. In this study, we developed AI-CURA, a framework that nearly fully automates genetic variant classification according to the American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) guidelines and Clinical Genome Resource (ClinGen) recommendations. The framework integrates evidence assessment for non-literature-based criteria, which can be automated using standard bioinformatic tools, with a separate LLM-supported assessment of literature-based evidence. Two state-of-the-art LLMs, DeepSeek-R1 and o3-mini-high, were tested for their performance in summarizing literature-derived evidence relevant to variant classification. We demonstrated that through careful prompt engineering and creation of ACMG-rule-specific knowledgebases, DeepSeek-R1 outperformed o3-mini-high and achieved high sensitivity and 100% specificity in interpreting ACMG rules that require understanding literature-based evidence. In further testing with 150 variants curated by ClinGen experts, DeepSeek-R1 showed high concordance with human curators in final diagnosis. Last, we showed that AI-CURA can also be used for classification reanalysis using 150 ClinVar variants with conflicting interpretations. Our study provides an LLM framework capable of automated variant classification in the diagnosis of genetic diseases and variant reanalysis.