Prediction of human missense variant effects from functional evidence.

Prediction of the functional impact of missense variants by training on functional assay data rather than on clinical labels or population frequencies.

Missense variant effect prediction remains a bottleneck in diagnostic genetics, and current predictors rely on clinical outcomes or population patterns rather than direct measures of functional impact, which creates data circularity. The authors present FuncVEP, a family of variant effect predictors trained on diverse functional data to predict the functional impact of missense variants. FuncVEP generalises across datasets and outperforms 48 existing predictors across a wide range of benchmarks, improving accuracy from 78.8% to 84.6% on functional benchmarks and from 90.1% to 92.4% on clinical benchmarks. Applied to the UK Biobank and the Mount Sinai Million Health Discoveries Program, it identified 210 new gene-phenotype associations involving 494 genes linked to inborn errors of immunity, with a discovery rate above that of state-of-the-art predictors.

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

The main point is not the accuracy gain, a modest 2.3 points on clinical benchmarks, but the break with circularity: a predictor trained on functional measurements does not re-learn ClinVar, which makes its use as computational evidence far more defensible before a classification committee. The decisive question for a laboratory remains the conversion of these scores into calibrated probabilities usable within ACMG frameworks, and whether performance holds for genes without high-throughput functional assays — that is, most of the genes interrogated by exome and genome sequencing.

Analysis by Dr Thibaut Benquey