Knowledge graph to dissect genotype phenotype associations
Résumé
Defining the mechanisms underlying complex traits and phenotypes requires understanding their genetic basis and how genes interact with environmental and lifestyle factors. Population-based prospective cohorts provide valuable resources for such research, collecting extensive phenotypic and omics data. However, the wide range of phenotypes in these cohorts makes it difficult to define homogeneous and/or clinically meaningful subgroups, limiting traditional genome-wide association studies (GWAS). We aim to develop a novel graph-based methodology to identify genotypephenotype associations. Our method represents data as a graph, where nodes correspond to variables (e.g., participants, phenotypes), and edges represent relationships between them. These edges can connect the same type of nodes (for example, participant-participant interactions) or different types (for example, bipartite interactions connecting participant to SNP). We applied our approach to the GOLD project, which includes comprehensive information on medical conditions, drug consumption, demographics, and genotypes of 10,000 participants. The graph contains four node types (participant, drug, SNP, phenotype) and 10 edge types, with attributes such as drug reimbursements, genotypes, and participant similarity. This knowledge graph representation allows us to use graph theory tools, including clustering algorithms, random walks, and deep learning-based graph representation methods. We expect to detect weak genotype-phenotype association signals that could not be detected by GWAS.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |