Mar 06, 2020 | 15:00—16:00
Abstract
In her master thesis, Christiane analysed datasets that describe specific phases in the development of mouse embryos. Methods from physics in combination with models of complex systems were used to gain new insights into the temporal structure of gene regulation during such developmental processes.
The data sets consist of about 400 cells and the RNA expression profiles of a few thousand of their genes. The data samples the embryonic development of mice capturing the transition from one cell type to another one.
The inference problem consists of two parts. The first part (i) is to infer a developmental order of cells in the sample. This part of the thesis is concerned with dimensional reduction and constructing a so-called pseudo-time description for the developmental process. The second part (ii) deals with inferring regulatory dynamics from the inferred pseudo temporal order of cells.
It is assumed assumed that a developmental process can be described by a linearized differential equation
𝑥̇𝑖=𝐽𝑖+𝐴𝑖𝑗𝑥𝑗
with a non-linear constraint forcing the species abundances x to remain positive. The goal is to find the parameters J and A that minimize the error we make applying the equation to the pseudo time ordered data, using the method of steepest descent.
Finally, both approaches were compared and genes that have the most impact on the process were identified.