Modeling Gene Regulation in Single-Cell Genomics: From Tensors for 3D Genome Architecture to Causal Gene Networks
Advances in single-cell technologies are enabling increasingly rich measurements of gene regulation, ranging from simultaneous profiling of 3D genome architecture and epigenomic features to large-scale CRISPR perturbation experiments coupled with singlecell transcriptomics. Yet, statistical analysis of such datasets remains challenging due to high dimensionality, complex dependence structures, and measurement limitations. This talk will present two complementary statistical frameworks for modeling gene regulation in single-cell genomics. Muscle, a semi-nonnegative joint tensor decomposition method, leverages simultaneously profiled single-cell Hi-C and DNA methylation data to uncover cell-type-specific patterns of chromatin conformation and epigenetic regulation with parameters aligned to interpretable genomic features. ARGEN addresses gene regulatory network inference from Perturb-seq data by developing a causal discovery and inference framework that recovers directed gene–gene relationships while remaining robust to arbitrary unobserved factors and gene omission. Together, these methods advance single cell genomic analysis by providing statistically principled tools for modeling gene regulation under heterogeneous data modalities and experimental constraints.
I am a Postdoctoral Researcher at the University of Pennsylvania, where I work with Professor Hongzhe Li. I received Ph.D. in Statistics from the University of Wisconsin-Madison in May 2025, where I was advised by Professor Sündüz Keleş. Before joining UW Madison, I earned a M.A. in Statistics and my B.A. in Economics and Statistics at Yonsei University, and studied Economics as an exchange student at Erasmus University Rotterdam. I aim to address problems related to understanding how genes are regulated by their corresponding regulatory elements. To tackle this challenge, I develop statistical methods for analyzing high-dimensional genomic data and integrating diverse genomic, epigenomic, and perturbational data to enhance statistical interpretability and rigor. My methodological interests center on (i) dimension reduction techniques, including tensor methods and multivariate response regression, and (ii) causal discovery and inference for heterogeneous data.
Hybrid, but in-person attendance is encouraged.
https://bit.ly/KwangmoonPark