Construire, exploiter et étendre un graphe de connaissances pour l’étude des liens entre métabolisme et santé

Summary

Metabolomics aims to determine metabolic profiles, describing the impact on a metabolism of an experimental condition or a particular phenotype. In human health, the study and comparison of these profiles are instrumental in the characterization of pathological mechanisms, toxic effects, or to identify new targets for diagnosis. The interpretation of these profiles requires the recontextualization of the biochemistry and, more broadly, the biology of the observations by integrating external knowledge from the scientific literature. From chemical databases to the literature, all this knowledge is accessible, but its volume and growth limit its exploration. In response to this information overload, new approaches need to be developed to help the researcher exploit this knowledge. Using Semantic Web technologies, the work presented in this thesis proposes the extraction and aggregation of a set of relationships between chemical compounds and biomedical concepts, to build an open knowledge graph named FORUM. The associative links between concepts are inferred from their co-mention frequency in the literature and augmented by the semantic representation of entities provided by chemical and biomedical vocabularies. Beyond its support for the interpretation of metabolic profiles, the potential of FORUM to suggest new hypotheses and explore relationships through the Web of data is also discussed. Despite its growth, the coverage in the literature on the metabolome remains partial. When interpreting metabolic profiles, the amount of available information for each observed compound can therefore be unequal. In order to extend the knowledge graph to metabolites neglected in the literature, a second approach combining network analysis and Bayesian statistics was also developed.
New suggested relationships with diseases are derived from the literature on the metabolic neighbors of the compound, and provide a new level of relationship in the knowledge graph 

Keywords : Metabolomics, Knowledge Graph, Ontologies, Knowledge Discovery, Bayesian statistics, Natural Language Processing
 

Contact

• Maxime Delmas (UMR Toxalim)
• Linkedin