Title: Robust differential variability testing for single-cell expression data: Bayes when single cells become big data
Authors: Catalina Vallejos - MRC Human Genetics Unit (United Kingdom) [presenting]
Abstract: Cell-to-cell transcriptional variability in seemingly homogeneous cell populations plays a crucial role in tissue function and development. Single-cell RNA sequencing (scRNAseq) can characterise this variability in a transcriptome-wide manner. However, scRNAseq is prone to high levels of technical noise, creating new challenges for identifying genes that show genuine heterogeneous expression. We introduce BASiCS - a high-dimensional Bayesian hierarchical framework which simultaneously performs data normalisation, technical noise quantification and downstream differential expression analyses, whilst propagating statistical uncertainty across these steps. Beyond traditional mean expression testing, BASiCS can robustly identify changes in variability between cell populations, providing novel insights in e.g. immune cell populations. However, BASiCS was implemented using a MCMC algorithm which is time-consuming, particularly when applied to the large datasets that are increasingly available in scRNAseq experiments. We illustrate how recent scalable variations of MCMC as well as approximate inference methods can provide improvements in computational efficiency, and the associated trade-off with estimation performance. Finally, we also discuss some of the general challenges involved in applying these methods to the high-dimensional (Bayesian) models that are often required to capture the multiple sources of variability that underlie high-throughput omics data.