UCSF

The three-dimensional (3D) organization of the human genome plays a crucial role in gene regulation, influencing interactions between genes and regulatory elements. Despite significant progress in genomics, the diversity of 3D chromatin contact patterns across human populations remains underexplored. This dissertation uses machine learning to predict 3D chromatin contact maps from genome sequences, revealing new insights into genome architecture among diverse populations. In Chapter 1, I provide a literature review and overview of human population and regulatory genetics in relationship to the 3D genome with a focus on machine learning techniques for studying each. In Chapter 2 I present the results of my study using a machine learning model to predict 3D genome for thousands of individuals, uncovering substantial 3D genomic diversity, particularly within African populations. It also identifies regions where 3D divergence occurs independently of sequence variation, especially in areas under low functional constraint. These findings underscore the importance of considering 3D genome organization in understanding gene regulation and its implications for health and disease.

This work demonstrates the utility of machine learning in exploring human genomic diversity, with significant implications for genomics, personalized medicine, and therapeutic development.