VarSight

VarSight prioritizes genomic variants for rare-disease genomic medicine by applying binary classification models that integrate variant annotations and phenotype information.

Key Features:

• Integration with Codicem: VarSight uses Codicem for pre-processing including variant annotation and filtering.
• Feature Utilization: The tool processes 95 features derived from genomic data and clinical annotations and selects the top 20 most informative features for classification.
• Classification Algorithms: VarSight employs binary classification algorithms that incorporate variant annotations and phenotype information to predict whether a variant is clinically reported.
• Phenotype-Based Gene Ranking: VarSight incorporates gene rankings based on Human Phenotype Ontology (HPO) terms using an internal cosine score and the external tool PyxisMap.

Scientific Applications:

• Rare disease variant prioritization: Prioritizes candidate pathogenic variants in genomic medicine for rare disease patients by integrating phenotype information with variant annotations to narrow candidate lists and improve diagnostic accuracy.
• Variant triage for manual curation: Ranks variants to facilitate efficient manual curation of clinically reported variants.

Methodology:

Pre-processing and annotation with Codicem; extraction of 95 genomic and clinical features with feature selection to the top 20; training of binary classification algorithms to predict clinically reported variants using variant annotations and phenotype information; incorporation of HPO-based gene rankings via an internal cosine score and PyxisMap; validation on a retrospective cohort of 237 Undiagnosed Diseases Network patients and benchmarking against four variant prioritization algorithms and two single-measure controls.