Machine Learning for Personalized Medicine

Statistical challenges in the analysis of single-cell transcriptomics data

by Catalina Vallejos

The revolution of transcriptomics - moving from bulk samples to single-cell resolution - can provide novel insights into tissue's function regulation. However, to analyse single-cell RNA sequencing (scRNA-seq) data, methods formulated for bulk experiments cannot be directly applied. In particular, normalisation strategies designed for bulk RNA-seq datasets can lead to unstable results at single-cell level. In addition, the effect of technical variation - reflected in weak correlations among technical replicates - is often ignored by such approaches. In this tutorial, illustrate BASiCS (Bayesian Analysis of Single-Cell Sequencing data) as an analysis tool for scRNA-seq datasets (Vallejos et al, 2015). BASiCS is a hierarchical Bayesian model where (i) normalisation, (ii) quantification of technical variation and (iii) a decomposition of the total variability of gene expression into technical and biological components are performed simultaneously, borrowing information from intrinsic transcripts and technical spike-in genes which are added to the lysis buffer and thence present at the same level in every cell. Adopting such integrated approaches is critical in this context, where an independent analysis of each modelling aspect can veil biological signal. BASiCS also provides an intuitive detection criterion for highly (or lowly) variable genes within the population of cells under study. This is formalised by means of tail posterior probabilities associated to high (or low) biological cell-to-cell variance contributions, quantities that have easy interpretation. We demonstrate BASiCS using gene expression measurements from mouse Embryonic Stem Cells.

>back to "Talks and Speakers"