IMPROVE-DD

IMPROVE-DD improves variant evaluation in developmental disorders by integrating multiple phenotype resources—including Human Phenotype Ontology (HPO) terms, sex, growth, and development data—with a Bayesian likelihood-ratio framework applied to genome-wide sequencing data to prioritize pathogenic variants.

Key Features:

• Integration of Diverse Phenotypic Data: Combines Human Phenotype Ontology (HPO) terms with sex, growth, and development metrics commonly available in clinical records.
• Bayesian Framework: Applies a Bayesian framework using likelihood ratios for both nominal and quantitative phenotypic data.
• Classifier for HPO Terms: Constructs classifiers for HPO terms that predict pathogenic variants across multiple genes.
• Reduction of False Positives: Integrates multiple phenotypic sources to reduce false positives compared with models based on single data types.
• Cross-Validation Performance: Demonstrates predictive performance during training with receiver operating characteristic results (AUC ≥ 0.6), indicating utility even with relatively small datasets.
• Genome-wide Sequencing Support: Operates on genome-wide sequencing data for variant prioritization.

Scientific Applications:

• Developmental disorder diagnosis: Supports diagnostic programs for genetically determined developmental disorders by integrating clinical phenotypic data to prioritize candidate variants.
• Variant pathogenicity discrimination: Aids clinicians in distinguishing pathogenic from non-pathogenic variants to improve diagnostic accuracy.
• Research and gene prioritization: Applied in research contexts such as the Deciphering Developmental Disorders (DDD) study, where evaluation of 77 genes with pathogenic/likely pathogenic variants showed integrated models outperform single-source models.

Methodology:

Data integration of HPO terms with sex, growth, and development metrics; Bayesian analysis applying likelihood ratios to nominal and quantitative data; and construction of classifiers for HPO terms and other phenotypic data.