Genetic variant pathogenicity prediction trained using disease-specific clinical sequencing data sets.

Area of Science:

• Genomics
• Clinical Genetics
• Bioinformatics

Background:

• Advances in DNA sequencing increase clinical genomic testing.
• Many variants are classified as unknown clinical significance due to limited evidence.
• Population-scale variant data can aid clinical interpretation.

Purpose of the Study:

• To compare genomic constraint metrics for missense variant pathogenicity prediction.
• To demonstrate the need for disease-specific pathogenicity classifiers.
• To develop and validate PathoPredictor, a novel disease-specific pathogenicity classifier.

Main Methods:

• Compiled a dataset from 17,071 patients undergoing clinical genomic sequencing for cardiomyopathy, epilepsy, or RASopathies.
• Compared various computational methods for predicting missense variant pathogenicity using regional variant constraint.
• Trained and evaluated PathoPredictor, an ensemble classifier integrating regional constraint and variant-level features.

Main Results:

• Established a novel clinical variant dataset for evaluating pathogenicity prediction methods.
• Demonstrated the superior performance of disease-specific classifiers.
• PathoPredictor achieved >90% average precision across 99 disease genes, with near-perfect accuracy for some.

Conclusions:

• Regional variant constraint is a valuable feature for pathogenicity prediction.
• Disease-specific classifiers significantly improve the accuracy of variant interpretation.
• PathoPredictor offers a robust tool for classifying missense variant pathogenicity in clinical settings.