Mots-Clés
deep learning
apprentissage profond
intelligence artificielle
artificial intelligence
metabolism
métabolisme
GNNs
diabète
diabetes
ontologies
pathways
omics
Description
Better understand metabolic diseases’ heterogeneity by integrating existing knowledge and multi-omics data with graph neural networks
Metabolic disorders such as obesity, type 2 diabetes (T2D) and metabolic fatty liver disease (MFLD) are a major burden on healthcare systems. They are often combined, with 80% of patients suffering from T2D being obese and 68% also suffering from MASLD, but this is not always the case. A better understanding of the underlying molecular networks and how they interact is crucial for better prevention and effective precision medicine.
We are looking for a PhD student who will integrate prior knowledge and multi-omics data with Graph neural networks (GNNs). GNNs make it possible to learn relationships between molecular partners represented by their nodes. The student will construct hybrid GNNs from ontologies, interactions, and pathway data, filtered by existing knowledge on metabolic disorders (e.g. polygenic scores, differential expression). The nodes will be updated via attention structures. These GNNs will be combined at the level of prediction heads.
The student will train them with multi-omics data from clinical cohorts comprising patients suffering from one or more metabolic diseases. The outputs will be, for example, the identification of specific subgraphs of subtypes of disorders or combinations of subtypes of disorders (e.g. subtype of MASLD leading to or accompanied by T2D), or the classification of nodes (genes) according to their importance/impact.
Finally, the student will study learned attention structures, in order to better understand the metabolic, signalling and gene regulation pathways involved in these diseases.