Recent advances in 3D genome mapping of single-nucleus and single-cell chromatin interactions have enabled the delineation of 3D genome features that cannot be captured by existing methods based on a population of cells. However, the development of computational methods that can effectively utilize these emerging, new 3D genome data types remains a major challenge. In this Ph.D. dissertation, I propose to develop a series of new algorithms based on graph representation learning for the analysis of 3D genome organization in single-nucleus and single-cell resolutions. The new hypergraph based architectures are expected to unveil multi-way chromatin interactions in single nuclei, cell-to-cell variability of 3D genome features, and single-cell connections between 3D chromatin structure and function. In addition, I propose a new framework by combining the graphical model and the graph neural network that jointly models chromatin interactions and one dimensional epigenomic features, providing a new paradigm for integrative analysis of multi-modal epigenomic data. Together, the new methods developed in this dissertation have the potential to provide key insights into the structure and function connections of 3D genome organization and will be of high value to a diverse group of biomedical researchers.
Jian Ma (Chair, CMU)
Ivet Bahar (Pitt)
Jure Leskovec (Stanford University)
Zoom Participation. See announcement.