MVP predicts the pathogenicity of missense variants by deep learning.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of missense variant pathogenicity is crucial for genetic research and clinical diagnostics.
• Existing prediction methods have limitations in performance and accuracy.

Purpose of the Study:

• To introduce MVP (Missense Variant Pathogenicity prediction), a novel deep residual network-based method for enhanced missense variant pathogenicity prediction.
• To improve the interpretation of missense variants, particularly in genes with varying loss-of-function intolerance.

Main Methods:

• Utilized a deep residual network architecture to leverage large datasets and numerous predictors.
• Trained the MVP model separately on genes intolerant and tolerant of loss-of-function variants.
• Benchmarked performance using cancer mutation hotspots and de novo variants from developmental disorders.

Main Results:

• MVP demonstrated superior performance in prioritizing pathogenic missense variants compared to previous methods.
• MVP showed particular effectiveness in genes tolerant of loss-of-function variants.
• MVP analysis estimated that de novo coding variants account for 7.8% of isolated congenital heart disease, nearly doubling prior estimates.

Conclusions:

• MVP offers a significant advancement in missense variant pathogenicity prediction.
• The findings highlight the substantial contribution of de novo coding variants to congenital heart disease, necessitating updated diagnostic and research approaches.