Using Networks and Prior Knowledge to Uncover novel Rare Disease Phenotypes

Rare diseases are characterized by low prevalence and high phenotypic diversity. Accurately identifying phenotypes associated with rare diseases is crucial for facilitating their diagnosis and management. However, this task presents significant challenges: rare disease datasets are typically small, making statistical assessments difficult, and they often report phenotypes using various terminologies, hindering the identification of common phenotypes across datasets or the recognition of those already documented in literature and knowledge databases.

The Xcelerate RARE 2023 challenge was established to address the identification of phenotypes associated with rare diseases. Our team, MAGNET, developed a network-based approach that integrates patient clinical data from the Xcelerate RARE 2023 challenge with existing knowledge from Orphanet and the Human Phenotype Ontology (HPO). Our approach first builds a patient-disease-phenotype network comprising two layers: the Xcelerate layer encoding disease-patient-symptom associations, and the Prior Knowledge layer incorporating relationships between Orphanet rare diseases and HPO phenotypes. Then, for each rare disease included in the Xcelerate dataset, a Random Walk with Restart (RWR) algorithm is applied to the multilayer network to prioritize phenotype nodes. This framework effectively prioritizes phenotypes associated with rare diseases while distinguishing novel phenotypes from those already documented in knowledge bases, hence offering new perspectives for improving the diagnosis and characterization of rare diseases.

Our solution was awarded the prize for the most innovative approach in the Xcelerate RARE challenge.