High dimensional tensor methods for multi-modal single cell genomics data

Statistical Bioinformatics Seminar

Speaker: Kwangmoon Park (University of Wisconsin-Madison)

Abstract: Emerging single cell technologies that simultaneously capture long-range interactions of genomic loci together with their DNA methylation levels are advancing our understanding of 3D genome structure and its interplay with the epigenome at the single cell level. While methods to analyze data from single cell high throughput chromatin conformation capture (scHi-C) experiments are maturing, methods that can jointly analyze multiple modalities with scHi-C data are lacking. In this talk, I present two tensor modeling frameworks: Muscle and SHOPS, to jointly analyze 3D conformation and DNA methylation data measured at the single cell level. First, I present Muscle, a joint decomposition of Multiple single cell tensors. Muscle is a novel tensor decomposition method that can integrate the scHi-C and DNA methylation modalities with a direct interpretability. Next, I introduce SHOPS, Sparse Higher Order Partial Least Squares, which provides an inference on the direct association between Hi-C and DNA methylation. SHOPS is a new tensor response regression method to simultaneously achieve denoising of the scHi-C tensor and selecting the most relevant methylation sites with dimension reduction.

About the speaker: Kwangmoon Park is a Statistics Ph.D. Candidate at the University of Wisconsin-Madison. He is currently working on statistical genomics and high dimensional statistics with Professor Sündüz Keleş. Before joining UW-Madison, he earned a master’s degree in Statistics at the Yonsei University in 2020. He earned a B.A. in Economics and Statistics at Yonsei University in Korea and studied Economics as an exchange student at Erasmus Universiteit Rotterdam in the Netherlands. Kwangmoon Park is mainly interested in questions related to understanding how genes are regulated by distal regions in the genome, particularly by functional non-coding regions. For that purpose, he develops statistical tools for analyzing High-dimensional genomic data, including Hi-C and HiChIP, and for linking diverse types of genomic or epigenomic data with better statistical interpretation. The statistical methodologies he works on are related to tensor factorization/regression and dimension reduction techniques, including Partial Least Squares.

Join on Zoom: https://uni-sydney.zoom.us/j/84087321707