Agenda

This thesis explores the biophysical and computational modeling of self-organization in
gastruloids, three-dimensional in vitro models of embryonic development. It investigates
how cellular decisions emerge from mechanogenetic interactions at the single-cell level,
leveraging advanced microscopy, image analysis, and tracking methodologies to quantify
cellular behaviors and micro-environmental cues.
The first part presents a methodological pipeline to quantify mechanical and genetic
properties in deep gastruloid tissues. A deep-learning-based segmentation strategy tailored
for heterogeneous nuclei shapes is then developed, integrating contrast enhancement and
dual-view fusion to improve imaging depth and accuracy.
Building on these segmentation advances, the study probes the mechanogenetic landscape
of gastruloids, integrating quantitative analyses of gene expression and cellular deforma-
tion. A dedicated Python-based computational pipeline is introduced, enabling multiscale
analysis and user-friendly visualization.
In the second part, the challenges of cell tracking in gastruloids are assessed, presenting
how stochasticity and measurement uncertainty complicate lineage reconstruction. Vari-
ous tracking methodologies applied in the context of cell tracking or developmental biology
are reviewed, leading to the development of a Bayesian filtering framework for the multi-
scale quantification of tracking uncertainty.
The final sections apply these probabilistic methods to validate segmentation and tracking
algorithms, demonstrating how the Bayesian filtering algorithm can guide unsupervised
parameter tuning and segmentation algorithm ranking. Through extensive simulations
and experimental applications, the study provides rigorous methods to aid cellular dynam-
ics analysis in complex gastruloid datasets, with implications for developmental biology
and tissue engineering.