Semantic embedding of variant effect annotations enables rapid and accurate pathogenicity prediction with VUS.Life
An AI tool accurately predicts whether genetic variants cause disease, potentially resolving the clinical bottleneck of thousands of genetic variants of uncertain significance.
VUS.Life is a multi-modal framework using semantic LLM text embeddings (MPNet, MedEmbed, text-embedding-004) + protein language modeling (ESMC-600M) for VUS pathogenicity classification, achieving high accuracy (MCC 0.895-0.989) across 8 ACMG Tier 1 genes including BRCA1/2, ATM, and PALB2. The approach is scalable, interpretable, and independent of existing pre-computed scores, offering a promising tool for alleviating the clinical VUS interpretation bottleneck.
What the study was
- Study design
- Computational tool development + cross-validation
- Population
- Variants of uncertain significance across 8 ACMG Tier 1 disease genes (BRCA1, BRCA2, FBN1, ATM, PALB2, MYH7, USH2A, PAH); n>10,000 variants
- Category
- Genomics/Precision Medicine
- Maturity
- Exploratory
- Journal
- Scientific Reports
Why it surfaced
Novel LLM-based approach to VUS classification with strong computational performance. Clinically relevant for genomic medicine pipeline, especially given millions of VUS in clinical databases. Scored 6: LOO-CV (not truly independent test set) and in silico only; clinical prospective validation needed.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.