Menu
Learning and inference of early human embryo morphogenesis

Learning and inference of early human embryo morphogenesis

By In PhD proposals 2019 On March 11, 2019


Project: Learning and inference of early human embryo morphogenesis

Laboratory: Multiscale Physics of Morphogenesis group – Center for Interdisciplinary Research in Biology

Affiliation
Affiliation: Collège de France, CNRS UMR7241, INSERM U1050
Address: 11, place Marcelin Berthelot, 75231 Paris Cedex 05
Website: www.virtual-embryo.com // www.college-de-france.fr/site/en-cirb/Turlier.htm
E-mail: herve.turlier@college-de-france.fr

LAB Director
Name: Hervé Turlier
Phone number: +33 (0)144271410
E-mail: herve.turlier@college-de-france.fr

Supervisor
Name: Hervé Turlier
Phone number: +33 (0)144271410
E-mail: herve.turlier@college-de-france.fr

Subject Keywords: human early embryo, morphogenesis, deep learning, inference, generative models
Summary of lab’s interests: Combining physical modeling and simulations, the lab studies the processes that shape biological systems across multiple scales. In the last years, our research has focused particularly on the surface forces controlling the shape of single cells (1) and the morphogenesis of multicellular embryos (2,3). Combining state-of-the-art techniques from various fields, the lab is developing a new simulation platform – Virtual Embryo – to analyze and simulate in 3D the morphogenesis of early embryos.

1. Turlier et al. Biophys J 2014
2. Maître, Turlier et al. Nature 2016
3. Dumortier et al. in revision 2019
Project summary: The project aims at learning and inferring the morphogenesis of early human embryos, from the zygote to the blastocyst, combining deep learning algorithms and physical modeling. While an ever-growing number of couples are facing fertility problems in modern societies, medical and biological research on early human embryos is understandably tightly regulated. Benefiting from a collaboration with a medical doctor, the goal of the project is to leverage the power of recent unsupervised machine learning algorithms to statistically infer the morphogenetic processes underlying preimplantation human development. By constraining a deep convolutional neural network with by a pre-defined physical model, we will extract the physically relevant parameters that characterize normal and abnormal development. Generalizing this approach to other mammalian embryos, such as mice, in collaboration with embryologists, we hope to be able to interpolate such representation from one specie to another by using recent generative models. This could provide alternative ways to test relevant hypothesis on human embryos by biologically mimicking its features in non-human embryos. This project will give us an unprecedented insight into human preimplantation development and could generate new perspectives on mammalian evolution.

Interdisciplinary aspect of the project: One of the cornerstones of this project is its interdisciplinary character, at the frontier between physics, data science and developmental biology. The project will involve bioimage analysis, physical modeling, numerical simulations and benefits from a collaboration with a team of embryologists and a medical doctor.

Funding: The position is not funded: the applicant will have to apply to PhD funding.
The project is yet supported by an interdisciplinary starting grant from the Convergence Institute Q-Life/PSL, obtained as a consortium, and providing computing ressources dedicated to this work.