AlphaMissense prediction for the evaluation of missense variants in the diagnostic setting of neuromuscular disorders.

Abstract

Next-generation sequencing has improved diagnostic outcomes for neuromuscular disorders, but interpreting rare missense variants remains challenging. We evaluated AlphaMissense, a recently developed machine learning tool, for predicting missense variant pathogenicity, using 45 (likely) pathogenic variants and 21 variants of uncertain significance from 58 deeply phenotyped patients. AlphaMissense predicted 69% of pathogenic variants correctly, but also classified 62% of variants of uncertain significance as pathogenic. Median AlphaMissense scores were not significantly different between pathogenic and uncertain variants. Overall, AlphaMissense accurately predicted the pathogenicity of most missense variants, but may be limited in certain functional contexts, highlighting the need for disease-specific interpretation approaches.