Ancestry-specific performance of variant effect predictors in clinical variant classification.

Area of Science:

• Genomic Medicine
• Computational Biology
• Human Genetics

Background:

• Variant effect predictors are crucial for genomic medicine, aiding in the genetic diagnosis of rare Mendelian conditions.
• Current predictors may exhibit performance disparities across genetic ancestries due to training data limitations.
• Responsible deployment of these tools necessitates understanding their ancestry-specific performance.

Purpose of the Study:

• To assess the ancestry-specific performance and accuracy of computational variant effect predictors.
• To evaluate the strength of evidence provided by predictors according to ACMG/AMP guidelines across different ancestries.
• To identify key factors influencing predictor performance across diverse populations.

Main Methods:

• Analyzed variant effect predictor performance stratified by genetic ancestry.
• Identified rare variant counts and allele frequency distributions as key confounding factors.
• Correlated predictor accuracy with the allele frequency of rare variants.

Main Results:

• Established methods for predicting missense variant pathogenicity demonstrate comparable performance across major ancestry groups when stratified by allele frequency.
• Predictor accuracy was found to be inversely correlated with the allele frequency of rare variants.
• Rare variant counts and allele frequency distributions significantly impact performance evaluation across ancestries.

Conclusions:

• Computational predictors for missense variant pathogenicity exhibit robust and comparable performance across major genetic ancestries.
• Allele frequency is a critical factor to consider when evaluating predictor performance.
• The findings support the widespread use of these predictive models in genetic diagnosis and related applications.