Title: Scalable clustering methods for dynamic Poisson graphical models
Authors: Yingying Wei - The Chinese University of Hong Kong (Hong Kong) [presenting]
old account LXY - The Chinese University of Hong Kong (Hong Kong)
Abstract: Graphical models describe the dependence structures among random variables. Recently there has been active research on modeling multiple Gaussian graphical models together. Nevertheless, there is a lack of research on Poisson Graphical models which describe counts data, let alone jointly modeling multiple Poisson Graphical models. We present a novel Bayesian nonparametric dynamic Poisson graphical model for multivariate counts data. The model was motivated by the transcription factors (TF) networks that control gene expression. The TF networks are dynamically varying across diverse biological conditions and heterogeneous across the genome within each given condition. Our proposed model automatically captures the between-condition dynamics and within-condition heterogeneity. Despite the motivating example, the proposed model is applicable to a large class of multivariate counts data. A parallel Markov Chain Monte Carlo algorithm is developed for posterior computation, which enables the computation to be scaled up efficiently.