PhenoSV

PhenoSV interprets and prioritizes structural variants to assess their phenotypic impacts and identify genes likely affected by those variants.

Key Features:

• SV types covered: Interprets all major structural variant types, including deletions, duplications, inversions, translocations, and other genomic rearrangements.
• Noncoding SV analysis: Analyzes noncoding SVs in addition to coding variants to assess regulatory and other noncoding impacts.
• Gene impact identification: Identifies genes affected by structural variants.
• Segmentation and annotation: Segments and annotates SVs using diverse genomic features.
• Phenotype-aware machine learning: Applies a phenotype-aware machine-learning model to integrate phenotype information into SV prioritization.
• Transformer architecture: Uses a transformer-based architecture within a multiple-instance learning framework to predict SV impacts.
• Benchmarking: Evaluated on human SV datasets covering all SV types and reported superior performance compared to existing methods.

Scientific Applications:

• Disease gene discovery: Prioritizes structural variants and identifies disease-related genes derived from SVs.
• Phenotype association: Integrates gene-phenotype associations to prioritize SVs likely related to specific phenotypes or diseases.
• Functional consequence assessment: Evaluates potential effects of SVs on gene function and expression.
• Genomic basis of phenotypic diversity: Supports research into the genomic underpinnings of phenotypic diversity and disease.

Methodology:

Segments and annotates SVs using diverse genomic features; employs a transformer-based architecture within a multiple-instance learning framework; and uses a phenotype-aware machine-learning model that integrates gene-phenotype associations to predict impacts on gene function and expression and to prioritize SVs.