Data-driven Computational Models of Fruit Fly Embryogenesis

Abstract

In the process of development, a single cell divides into multiple cells, differentiates into tissues and ultimately the tissues go through mechanical deformations to give rise to functional organs. The processes have been studied with increasingly sophisticated tools of genetics, molecular biology, and microscopy in the early fruit fly (Drosophila melanogaster) embryo. In this thesis, we show how computational analysis of the experimental data coupled with biophysical models can lead to mechanistic understanding and predictive power about the processes of development. In the first part of the thesis (chapter 2), we focus on the process of nuclear divisions in the embryo. We extracted point patterns formed by centers of the nuclei in successive nuclear cycles analyzing the live imaging data. Descriptors from statistical mechanics revealed that while the point pattern doesn’t have any long-range order, the characteristic distance between them scales with inverse square-root of density as the number density of the nuclei doubles with mitotic divisions. We showed that a particle-based model with adaptive inter-particle force-field reproduces this feature and used that model to construct a virtual embryo on the surface of a prolate spheroid with nuclear positions similar to the real embryo. In Chapter 3, we focus on gene regulation by signaling network, the primary mechanism used by the embryo for cellular differentiation. We used experimental data from different levels in the signaling pathway for the terminal patterning system of fruit fly controlled by RAS/ERK signaling. We proposed an integrated model that takes into multiple events such as nuclear division, shuttling of transcription factor into nucleus, binding of transcription factor to DNA, and transcription and learned the parameters of the model from the experimental data that makes quantitative prediction and gives us mechanistic understanding. Next, we utilized similar models on the virtual embryo generated from chapter 2 to model gene expression patterns with a single nucleus resolution. In the final part of the thesis (chapter 4), we focus on embryo-scale movements of the cells that lead to morphogenetic deformation. We integrated live imaging, tissue cartography, and particle image velocimetry to construct the time dependent velocity field on the surface of the virtual embryo. Finally, we used dimensionality reduction techniques to show that wild type embryos are characterized by stereotype modes in space and time, whereas, embryos which are metabolically perturbed show self-organized oscillatory instability. Altogether, this thesis tries to integrate a diverse set of data analysis and modeling techniques to integrate biological understanding obtained over decades to a quantitative form with predictive power. Furthermore, this thesis is also a first step towards integrated self-consistent models of development that integrate multiple stages and events.