Functional effect predictions for ion channel missense variants using a protein language model

Protein language model trained on 1,996 electrophysiologically characterized variants to predict functional effects (gain or loss of function) of missense variants in 600+ ion channel genes

A protein language model is trained on the largest dataset of electrophysiologically characterized ion channel variants (1,996 variants), enabling prediction of functional effects (gain or loss of function) of missense variants in more than 600 ion channel genes. The tool achieves an AUC-ROC of 0.918 versus 0.884 and 0.779 for competing models, and generalizes to genes not represented in training. An online interface makes predictions accessible for 600,000+ variants.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Channelopathies (SCN, KCNQ, CACNA...) represent a major cause of hereditary cardiac and neurological diseases with a significant VUS burden. A gain/loss-of-function predictor specific to ion channels with AUC >0.90 is a directly applicable advance for VUS classification in these genes.