Prediction of human missense variant effects from functional evidence

Abstract Prediction of missense variant effects remains the critical bottleneck in disease gene identification and clinical interpretation. Current predictors rely on clinical outcomes or population patterns, rather than direct measures of functional impact, leading to limited generalizability and data circularity. We present FuncVEP, the first family of variant effect predictors trained exclusively on balanced and diverse functional data, providing a direct representation of functional effect. FuncVEP generalizes across contexts, outperforming 47 existing predictors on both clinical and functional benchmarks, improving the accuracy from 82% to 93% and reducing uncertain classifications from 11% to 2%. To illustrate its utility in gene discovery, we applied FuncVEP to 490 inborn errors of immunity genes in the UK Biobank and Mount Sinai Million Health Discoveries Program, identifying 50 novel gene–phenotype associations. FuncVEP provides a robust, scalable solution for variant interpretation, advancing both diagnostic precision and gene discovery.