Cytocipherdetermines significantly different populations of cells in single cell RNA-seq data

Abstract

Identification of novel and known cell types with single cell RNA-seq (scRNA-seq) is revolutionising the study of multicellular organisms. However, typical scRNA-seq analysis involves many preprocessing steps and represents an abstraction of the original measurements, often resulting in clusters of single cells that may not display distinct gene expression. To mitigate this, cell clusters are typically validated as cell types by re-examination of the original expression measurements, often resulting in post-hoc manual alteration of clusters to ensure distinct gene expression. However, distinct cell populations may exist that are not clearly demarcated by a single marker gene, but instead co-express a unique combination of genes; a phenomenon which is difficult to detect by manual examination. Furthermore, manual examination of genes and post-hoc cluster editing is time-consuming, error-prone, and irreproducible. Here, we present Cytocipher, an scverse compatible bioinformatics method and software that scores cells for unique combinatorial gene co-expression and statistically tests whether clusters are significantly different. Application to both simulated and real data demonstrates that the combinatorial gene expression scoring outperforms existing per-cell gene enrichment methods, such as Giotto Parametric Analysis of Gene Set Enrichment and Scanpy-score. Furthermore, Cytocipher cluster-merging identified distinct CD8+ T cell subtypes in human peripheral blood mononuclear cells that were not identified in the original annotations. Cytocipher cluster-merging was also able to identify distinct intermediate states corresponding to cell lineage decisions and branch points in mouse pancreas development, not previously identified in the original annotations. Identification of significantly different clusters of cells is an important new methodological improvement to the existing analysis pipeline of scRNA-seq data. Utilisation of Cytocipher will thus ensure that single cell Atlas mapping efforts provide distinctly different and programmatically reproducible cell clusters. Cytocipher is available at https://github.com/BradBalderson/Cytocipher.

CHOIR improves significance-based detection of cell types and states from single-cell data

“…Furthermore, using published cell type annotations as a stand-in for ground truth cell identities risks biasing benchmarking results in favor of methods similar to those used in the original analysis. We therefore used complementary strategies in a three-tiered benchmarking approach to comprehensively compare the performance of CHOIR with that of 14 other unsupervised clustering methods 5,6,9,10,[16][17][18][19][20][21][22][23][24][25] (Supplementary Table 3) in the analysis of both simulated and real data (Supplementary Fig. 1c and Supplementary Table 1).…”