Unravelling cell fate decisions through single cell methods and mathematical models

Despite being the object of intense study, embryonic development has been difficult to model due to a number of reasons. First, complex tissues can be comprised of many cell types, of which we probably only know a subset. Therefore, we first focused on the discovery of cell types by single-cell RNA-sequencing (scRNA-seq). Cell types are routinely identified by clustering scRNA-seq data, however, there was no principled way to determine the right number of clusters. To improve cell type classification, we developed phiclust, a clusterability measure for scRNA-seq. Another challenge in a developing tissue is that many signaling processes and morphogenic events occur simultaneously, which makes it hard to isolate the individual contributions. For this purpose, I looked at stem cell derived in vitro systems, in which a small number of specific cell types can be combined deliberately and studied in isolation. My analysis of different model systems shows that cellular communication causes structural and transcriptional changes in the developing cells. Finally, while tissue organization has been characterized extensively, we lack generative models that can relate specific patterns to the underlying gene regulatory mechanisms. Therefore, I later focused on deep learning-based approaches to infer gene regulatory networks from observed spatial patterns. Biological and Soft Matter Physic