PERADIGM: Phenotype embedding similarity-based rare disease gene mapping.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Rare disease gene identification is challenging due to limited patient data and statistical power.
• Traditional methods often rely on binary disease status, limiting nuanced phenotype analysis.

Purpose of the Study:

• To introduce PERADIGM, a novel framework for rare disease gene discovery.
• To leverage natural language processing (NLP) and phenotype similarity for enhanced gene mapping.

Main Methods:

• PERADIGM utilizes an embedding model to represent relationships between ICD-10 codes, capturing nuanced individual phenotypes.
• Patient similarity scores are employed to improve the identification of candidate genes associated with specific rare disease phenotypes.
• The framework was applied to the UK Biobank dataset for autosomal dominant polycystic kidney disease (ADPKD), Marfan syndrome, and neurofibromatosis type 1 (NF1).

Main Results:

• PERADIGM identified additional candidate genes for ADPKD and Marfan syndrome phenotypes, with some findings supported by existing literature.
• The framework demonstrated enhanced signal detection for NF1-specific phenotypes compared to traditional methods.
• The study successfully integrated phenotype embeddings and patient similarity for rare disease gene discovery.

Conclusions:

• PERADIGM offers a powerful tool for identifying genes associated with rare diseases and their related phenotypes.
• The framework enhances gene discovery by incorporating phenotype embeddings and patient similarity, advancing precision medicine.
• This approach deepens the understanding of rare disease genetics and clinical manifestations.