Particle.news
Download on the App Store
3 ARTICLES
·
last mo.

AI Model popEVE Delivers Proteome-Wide Severity Scores for Rare-Disease Diagnosis

By fusing evolutionary patterns with a protein language model plus UK Biobank and gnomAD variation, the tool produces calibrated cross-gene scores that curb ancestry bias.

Image

Overview

  • Published in Nature Genetics, popEVE ranks missense variants across roughly 20,000 human proteins to pinpoint likely disease-causing changes.
  • In validation on more than 31,000 families with severe developmental disorders, it correctly prioritized the known causal variant in 98% of cases and outperformed AlphaMissense.
  • Applied to about 30,000 undiagnosed patients with severe developmental disorders, the approach yielded diagnoses in roughly one-third of cases.
  • The study nominated 123 previously unlinked candidate genes for developmental disorders, with about 25 subsequently confirmed by independent labs.
  • Real-world testing is underway through an online portal, collaborations with Boston Children’s, CHOP, and Genomics England/Wellcome Sanger, and integration of scores into ProtVar and UniProt.