HubViz: A Statistical Model for Hub-Centric Visualization of Multivariate Binary Data

• Activity Number: 124 - Recent Advances in Network Modeling and Visualizations
• Type: Topic Contributed
• Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Korean International Statistical Society
• Author(s): Dongjun Chung* and Jin Hyun Nam and Ick Hoon Jin
• Companies: The Ohio State University and Medical University of South Carolina and Yonsei University
• Keywords: Visualization; Latent space modeling; Bayesian estimation

Abstract

Visualization algorithms have been widely used for intuitive interrogation of genomic data. For example, multidimensional scaling (MDS) has been popularly used for visualization for multiple decades. Recently t-SNE and UMAP gain a lot of attention as they were proven to be powerful for visualization of single-cell genomic data. While binary data is ubiquitous in biomedical research and also many genomic data can be dichotomized, currently popularly used visualization methods are not tuned for visualization of binary data and none of them consider the hubness of observations for visualization. In this presentation, I will discuss hubViz, our novel statistical model for the hub-centric visualization of multivariate binary data. By adopting a latent space joint modeling approach, our proposed model can estimate pairwise distances to represent the dependence structure, from which clusters in latent spaces can be identified. I will illustrate hubViz with simulation studies and its application to the single-cell and bulk gene expression data, and the text mining data.