Prediction of human missense variant effects from functional evidence

Area of Science:

• Genetics
• Bioinformatics
• Genomic Medicine

Background:

• Predicting missense variant effects is crucial for identifying disease genes and interpreting clinical results.
• Existing prediction methods often rely on indirect data, limiting their accuracy and generalizability.
• A need exists for variant effect predictors trained on direct functional impact data.

Purpose of the Study:

• To introduce FuncVEP, a novel variant effect predictor family.
• To demonstrate FuncVEP's superior performance compared to existing predictors.
• To showcase FuncVEP's utility in identifying novel gene-phenotype associations.

Main Methods:

• Developed FuncVEP, a variant effect predictor trained exclusively on balanced and diverse functional data.
• Evaluated FuncVEP against 47 existing predictors using clinical and functional benchmarks.
• Applied FuncVEP to inborn errors of immunity genes in large-scale datasets (UK Biobank, Million Health Discoveries Program).

Main Results:

• FuncVEP achieved a higher accuracy of 93%, an improvement from 82% over existing predictors.
• Uncertain variant classifications were reduced from 11% to 2%.
• Identified 50 novel gene-phenotype associations in inborn errors of immunity.

Conclusions:

• FuncVEP offers a robust and scalable solution for missense variant interpretation.
• The predictor enhances diagnostic precision and accelerates gene discovery.
• FuncVEP's reliance on direct functional data improves generalizability and reduces data circularity.