PhenoScore quantifies phenotypic variation for rare genetic diseases by combining facial analysis with other clinical...

Area of Science:

• Genomics
• Bioinformatics
• Artificial Intelligence

Background:

• Existing genotype-phenotype correlation methods lack a unified framework integrating facial and other phenotypic data.
• Facial recognition and Human Phenotype Ontology (HPO) data analysis are valuable but often used separately.

Purpose of the Study:

• To develop and validate PhenoScore, an open-source AI-driven phenomics framework for quantifying phenotypic similarity.
• To establish a unified approach for genotype-phenotype correlation using both facial and non-facial phenotypic data.

Main Methods:

• Development of PhenoScore, an AI framework combining facial recognition with HPO data analysis.
• Quantification of phenotypic similarity and recognition of distinct phenotypic entities.
• Validation against 40 investigated syndromes and individuals with neurodevelopmental disorders.

Main Results:

• PhenoScore successfully recognized recognizable phenotypes for 37 of 40 investigated syndromes.
• The framework demonstrated improvement over existing genotype-phenotype correlation approaches.
• PhenoScore identified distinct phenotypic subgroups for SATB1, SETBP1, DEAF1, and ADNP-related disorders.

Conclusions:

• PhenoScore offers a novel, unified framework for genotype-phenotype correlation by integrating diverse phenotypic data.
• The AI-based tool enhances the identification and characterization of genetic syndromes, including neurodevelopmental disorders.
• PhenoScore aids in predicting outcomes for individuals with variants of unknown significance and facilitates sophisticated genotype-phenotype studies.