A clustering-independent method for finding differentially expressed genes in single-cell transcriptome data

Abstract

A common analysis of single-cell sequencing data includes clustering of cells and identifying differentially expressed genes (DEGs). How cell clusters are defined has important consequences for downstream analyses and the interpretation of results, but is often not straightforward. To address this difficulty, we present singleCellHaystack, a method that enables the prediction of DEGs without relying on explicit clustering of cells. Our method uses Kullback-Leibler divergence to find genes that are expressed in subsets of cells that are non-randomly positioned in a multidimensional space. Comparisons with existing DEG prediction approaches on artificial datasets show that singleCellHaystack has higher accuracy. We illustrate the usage of singleCellHaystack through applications on 136 real transcriptome datasets and a spatial transcriptomics dataset. We demonstrate that our method is a fast and accurate approach for DEG prediction in single-cell data. singleCellHaystack is implemented as an R package and is available from CRAN and GitHub.

Building the Mega Single Cell Transcriptome Ocular Meta-Atlas

“…As another check on the quality of the cell type assignments, we ran the cell and cluster independent haystack gene search and pairwise differential expression tests between the predicted cell types (see methods for further details). 47 We show the five most differentially expressed genes for each of the major retina cell types are consistent with known retinal cell markers (Figure 3a). As a simple metric to identify known and unknown genes relating to the cell type specific expression we search PubMed for the number of publications with two searches per gene.…”

“…To identify marker genes across the CellType (predict) and cluster groups, we used the scran findmarkers (wilcox test) along with the singleCellHaystack algorithm. 47 The scran findmarkers test runs a wilcox test in a pairwise manner (e.g. Rods vs all other cell types).…”

Exploring tissue architecture using spatial transcriptomics

“…Gene selection methods abound, and those tailored to spatial transcriptomic data attempt to identify genes with high variance and whose expression is not random across the tissue. Genes can be scored according to their spatial autocorrelation (using Moran's I or Geary's C) 125 , neighbor enrichment (for example, in BinSpect) 111 or entropy (for example, in Haystack) 126 . Trendsceek 127 uses a marked point processes approach 128 and is able to identify hotspots, streaks, and gradients of expression.…”

MarcoPolo: a clustering-free approach to the exploration of differentially expressed genes along with group information in single-cell RNA-seq data

“…As the differential expression analysis in downstream depends on the clustering result, if a clustering algorithm fails to properly cluster a group of cells, the informative genes of that cluster will be missed. In this sense, the need for a method that can find DEGs independent of clustering was widely discussed 5,13,14 . We demonstrated this situation as a proof-of-concept using realistic simulation datasets generated by Symsim 19 , a simulator of single-cell RNA-seq experiments.…”

“…In our analysis, we show that they do not assign higher ranks to DEGs that are obviously differentially expressed among cell types with sufficient precision compared to the standard DEG analysis with clustering. Another more advanced approach is singleCellHaystack, a recently developed method that extracts the list of candidates for DEGs by examining non-random expression pattern before clustering 7 . However, although the method may sort out DEGs more accurately than the HVG methods, it does not tell which groups of cells ‘differentially’ express the identified genes.…”