Co-expression in Single-Cell Analysis: Saving Grace or Original Sin?

Abstract

As a fundamental unit of life, the cell has rightfully been the subject of intense investigation throughout the history of biology. Technical innovations now make it possible to assay cellular features at genomic scale, yielding breakthroughs in our understanding of the molecular organization of tissues, and even whole organisms. As these data accumulate we will soon be faced with a new challenge: making sense of the plethora of results. Early investigations into the replicability of cell type profiles inferred from single-cell RNA sequencing data have indicated that this is likely to be surprisingly straightforward due to consistent gene co-expression. In this opinion article we discuss the evidence for this claim and its implications for interpreting cell type-specific gene expression.

Accurate denoising of single-cell RNA-Seq data using unbiased principal component analysis

“…To develop a strategy for denoising scRNA-Seq datasets containing a complex mixture of cell populations with unknown degrees of similarity, we observed that while true expression differences between cell types and subpopulations result in substantial gene-gene correlation structure 22,23 , sampling noise affects the expression measurement of each gene in a largely independent fashion. We therefore reasoned that principal component analysis (PCA), applied to variance-stabilized data 10 , could be a powerful approach for separating biological heterogeneity from technical noise in scRNA-Seq data.…”

Coexpression enables multi-study cellular trajectories of development and disease

“…As baselines, we computed (1) (Figure 5a,b). We reasoned that this result is due to more discoverable gene-gene interactions 284 (as captured by coexpression) within the pan-resolution setting because coexpression changes in 285 strength with clustering resolution 16,24 . We also note that this result is not limited to the multi-286 study integration setting but can, in principle, also increase discovery of coexpressed genes 287 within a single study.…”

“…Moreover, biological signal in multi-study analysis is 13 confounded by study-specific noise patterns. This problem has motivated techniques for 14 computational batch effect correction [9][10][11][12][13][14][15] , but existing approaches integrate experiments using 15 transformations that obscure the biological relevance of individual data values, making it 16 difficult for downstream analyses to interpret the transformed result. Existing integrative 17 algorithms also aim to minimize inter-study variation, thus removing relevant differences that 18 would otherwise be useful to biological researchers.…”

“…measurements; not only are many coexpression measures (for example, Pearson correlation) 25 robust to affine transformation, some evidence suggests that gene coexpression and information 26 redundancy underlie cross-study replicability of single-cell experiments [16][17][18] . Coexpression also 27 provides a rich feature space with directly meaningful values that capture pairwise dependencies 28 among genes, allowing for graph-theoretic analysis of gene coexpression networks.…”

“…Coexpression is typically measured by computing a 53 gene-by-gene correlation matrix over a cluster of cells. Coexpression measurements, however, 54 may change with clustering resolution 16,24 and single cell datasets often have meaningful 55 multiresolution structure 25 . We therefore introduce a strategy that repeatedly clusters a dataset at 56 multiple resolutions and considers all clusters for downstream analysis (Figure 1a ; Methods); 57 we refer to this strategy as pan-resolution clustering (panclustering).…”