Organoid Phenotypes Mapping and Modeling : Toward an Endocrine Disruptors Classification – MORPHEUS

Organoid are a recent technology very promising in different medical applications(Characterization of molecules effect, drug choice for personalized medicine). Computational tools are now required for fully taking benefit from this approach. With this respect, this project aims at filling this gap in the case of endocrinian disruptors effect understanding. These endocrinian disruptors are of a major concern in european community for taking decisions, notably in terms of agriculture management. This project has two main aims. The first one consists in deriving machine learning tools to draw a phenotypic map of endocrinian perturbators.

To achieve this goal we will first consider deep learning approaches for their discriminative preperties by developing an neural network for classifying the different phenotypes. The classes will be represented in the bottleneck space. To improve the classes interpretability and to better characterize the geometrical and topological preperties of the phenotypes we will define a space of graphs as a stratifold where each point represents a given organoid by its graph (the nodes being the cells and edges defining adjacency between cells). We then will define a mapping between the "bottleneck" and the graph spaces.

The second goal concerns organoid growth modeling at the local and global scales. This modelling will rely on the obtained phenotypic map. The growth of a given organoid will be assoicated to a trajectory in this space. A classification algorithm will allow grouping the samples with respect to their phenotypic dynamic during growth. The effects of endocrinien disruptors will be characterized by the deviations from the trajectories.

Experiments will be done on two models whcih are the prostate organoids and the gastruloids. These two models will ensure several different phenotypes. These claases will be enriched due to the adaptative property of the proposed approach by filling the description of the two definend spaces.