StrVCTVRE

StrVCTVRE predicts the pathogenicity of structural variants (SVs), specifically deletions and duplications overlapping exons larger than 50 base pairs, to prioritize likely disease-causing variants.

Key Features:

• Pathogenicity Prediction: Predicts whether exon-overlapping deletions and duplications are pathogenic or benign.
• Random Forest Classifier: Uses a random forest classifier that integrates gene importance, coding-region characteristics, conservation levels, expression data, and exon structure.
• Feature Integration: Incorporates expression and conservation alongside gene and exon features, which are absent from current SV classification guidelines.
• Training Set Construction: Builds a size-matched training set of rare, putatively benign and pathogenic SVs compiled from multiple resources.
• Performance Metrics: Shows high accuracy across a wide range of SV sizes on independent test sets, reducing the number of SVs requiring further consideration by about half while maintaining 90% sensitivity.

Scientific Applications:

• Clinical Prioritization: Prioritizes SVs in probands lacking an immediately compelling single variant to aid case resolution.
• Long-read Sequencing Diagnostics: Assesses exon-overlapping SV pathogenicity to maximize diagnostic yield from long-read sequencing.

Methodology:

Implements a random forest classifier that integrates gene importance, coding-region features, conservation, expression, and exon structure; trains on a size-matched set of rare putatively benign and pathogenic SVs compiled from multiple resources; evaluates performance on independent test sets.