Optimizing Detection Techniques for Genomic Single-Molecule Localization Microscopy Point-Cloud Data: A Multi-Chromosome Application

The three-dimensional (3D) organization of chromosomes within the cell nucleus is essential for regulating gene expression, maintaining genome integrity, and supporting proper cellular function. In this thesis, we investigate chromatin structure using data generated by OligoSTORM, a single-molecule localization microscopy (SMLM) technique that integrates Oligopaint fluorescence in situ hybridization (FISH) with stochastic optical reconstruction microscopy (STORM). This imaging approach enables nanometer-scale visualization of specifically targeted genomic regions and produces high-resolution point clouds representing the spatial positions of individual fluorescently labeled DNA molecules. The work presented here focuses on developing a data-driven computational framework for processing and analyzing dense point clouds OligoSTORM data, integrating machine learning and optimization techniques. The framework addresses various pivotal steps, ranging from detecting signals from noise to reconstructing the chromatin path from multiplexed experiments. Firstly, we implemented an approach for isolating signal from noise and segmenting relevant chromatin regions through a combined clustering and classification strategy. This approach relies on data-based decision processes and includes an adaptive deep learning component capable of improving its performance as new experimental data become available. Secondly, we also implement an optimization-based path-fitting procedure to refine the inferred chromatin path and ensure the spatial coherence of the reconstructed structures. This stage contributes to the separation of homologous chromosomes, enabling the interpretation of the 3D genomic organization. Overall, this work presents a modular and extensible computational pipeline designed for large-scale chromatin imaging data. Its data-driven and adaptive nature ensures that the framework can be readily maintained and incrementally improved as new clustered datasets and annotations are produced, offering a scalable solution for future research in computational genomics and spatial nuclear organization. To show the usability of the framework we applied on a multi-chromosome dataset involving chromosomes 22 and 9, where the objective is to reconstruct and interpret the structural arrangement of chromatin within the nucleus.