Spatial organization of DNA across scales: a Bayesian inference and information-theoretic approach

Spatial organization plays a fundamental role in many biological processes, from the folding of chromatin within the nucleus and the patterns of gene expression levels that guide cell fate decisions within an embryo during development. Understanding how the spatial arrangement of genetic material is connected to biochemical processes and cellular function is crucial for deciphering mechanisms of gene expression and development. This thesis presents an investigation across multiple scales of these relationships, focusing first on chromatin organization and nuclear biochemistry, and then extending to the properties of gene expression patterns that guide tissue development. In Chapter 2, SEMPER, a polymer-based model, is introduced to quantify the role of chromatin’s polymeric nature in shaping biochemical landscapes. By integrating one-dimensional biochemical data, such as transcription factor binding and histone modifications, into a three-dimensional linear scaffold, SEMPER highlights the influence of chromatin’s physical structure on epigenetic patterning. Chapter 3 shifts perspective and approaches the problem from a complementary angle by introducing bioSBM, a stochastic block model designed to predict chromatin structure starting from epigenetic features. This model provides an interpretable and generalizable framework for linking biochemical data to large-scale chromatin folding. BioSBM successfully predicts chromatin interactions across different cell types and conditions, suggesting that the relationships between epigenetic marks and chromatin structure are broadly conserved. In Chapter 4, the thesis expands to the tissue level, examining how spatial gene expression patterns along embryonic axes guide cellular differentiation and tissue organization. The concept of correlational information (CI) is introduced to quantify the degree of spatial correlations in gene expression fluctuations and is shown to help cells improve their positional accuracy during development. A non-local decoding framework is proposed, suggesting that cells can leverage information from neighboring cells to refine their own positional decisions.