Measuring cell-to-cell expression variability in single-cell RNA-sequencing data: a comparative analysis and applications to B cell ageing

BackgroundSingle-cell RNA-sequencing (scRNA-seq) technologies enable the capture of gene expression heterogeneity and consequently cell-to-cell variability at the cell type level. Although different methods have been proposed to quantify cell-to-cell variability, it is unclear what the optimal statistical approach is, especially in light of challenging data structures that are unique to scRNA-seq data like zero inflation. ResultsIn this study, we conducted a systematic evaluation of cell-to-cell gene expression variability using 14 different variability metrics that are commonly applied to transcriptomic data. Performance was evaluated with respect to data-specific features like sparsity and sequencing platform, biological properties like gene length, and the ability to recapitulate true levels of variability based on simulation and known biological gene sets like ribosomal genes and stably expressed genes. scran had the strongest all-round performance, and this metric was then applied to investigate the changes in cell-to-cell variability that occur during ageing. Studying ageing showcases the value of cell-to-cell variability as it is a genetically-regulated program that is influenced by stochastic processes.scRNA-seq datasets from hematopoietic stem cells (HSCs) and B lymphocytes and other cell types from this differentiation lineage were used with scran to identify the genes with consistent patterns of variable and stable expression profiles during differentiation. Furthermore, to understand the regulatory relationship for genes that were differentially-variable in their expression between young and old mice, we constructed networks using transcription factors and their known targets for HSC and B lymphocyte cells. Comparisons of these networks identified a shared TF Sfpi1 that although was seen to increase in gene expression variability in old mice versus young in both cell types, the corresponding targets were distinct and their gene expression variability had different directions between cell types. ConclusionsThrough these analyses, we highlight the importance of capturing cell-to-cell gene expression variability in a complex biological process like differentiation and ageing, and emphasise the value and specificity of interpreting these findings at the level of individual cell types.