B1796
Title: Bayesian single-cell transcriptome analysis and related MCMC convergence diagnostics
Authors: Sheng Lian - The Chinese University of Hong Kong (China) [presenting]
Abstract: Single-cell RNA sequencing has become widely used in recent years. The ability to measure the gene expression at a cellular resolution allows us to study new biological questions related to cell-specific changes. However, there exists a large amount of technical noise that may hinder downstream analysis. Dropout events happen when a gene is not detected owing to a failure to capture or amplify. Based on the assumption that the detection rate for each gene in every single cell depends on the level of expression, we propose a hierarchical model with the non-ignorable missing-data mechanism to model the dropout events. Bayesian inference based on Markov Chain Monte Carlo (MCMC) algorithms is performed. Also, trace plot for the posterior of the joint parameter is used for diagnosing convergence problems and some unexpected behaviors are observed. Then, we investigate the reasons in hierarchical modeling framework and provide guidance in MCMC convergence diagnostics. Finally, we move into single-cell cancer genomic studies and work on the method to identify differentially expressed genes between normal and diseased conditions. We conduct statistical analysis that accounts for the dropout events and illustrate how this step help us better understand disease heterogeneity at the single-cell level.
