muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data.
Crowell HL, Soneson C, Germain PL, Calini D, Collin L, Raposo C, Malhotra D, Robinson MD. · 2020
Read muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data. by Crowell HL, Soneson C, Germain PL, Calini D, Collin L, Raposo C, Malhotra D, Robinson MD. (2020) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc7705760.txt.
Introduction
Crowell and co-authors sort out how to test gene shifts across cell groups and conditions. Single-cell studies now span many samples and treatments at once. The question is which cell types change state between conditions. Standard tests built for single samples can mislead here, since cells from one sample are not independent.
The team compared cell-level models with methods that pool cells into pseudobulk samples. They built a flexible simulation that mimics real multi-sample data. They also applied the tools to mouse cortex cells after a bacterial toxin challenge. Their findings ship as the muscat package, with guidance on robust analysis.
Catalog record
| CREATOR | Crowell HL, Soneson C, Germain PL, Calini D, Collin L, Raposo C, Malhotra D, Robinson MD. |
|---|---|
| YEAR | 2020 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 80 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
muscat detects subpopulation-specific state transitions from multi-sample multi-condition single-cell transcriptomics data By Helena L. Crowell, Charlotte Soneson, Pierre-Luc Germain, Daniela Calini, Ludovic Collin, Catarina Raposo, Dheeraj Malhotra, Mark D. Robinson Published in: Nature Communications DOI: 10.1038/s41467-020-19894-4 ABSTRACT Single-cell RNA sequencing (scRNA-seq) has become an empowering technology to profile the transcriptomes of individual cells on a large scale. Early analyses of differential expression have aimed at identifying differences between subpopulations to identify subpopulation markers. More generally, such methods compare expression levels across sets of cells, thus leading to cross-condition analyses. Given the emergence of replicated multi-condition scRNA-seq datasets, an area of increasing focus is making sample-level inferences, termed here as differential state analysis; however, it is not clear which statistical framework best handles this situation. Here, we surveyed methods to perform cross-condition differential state analyses, including cell-level mixed models and methods based on aggregated pseudobulk data. To evaluate method performance, we developed a flexible simulation that mimics multi-sample scRNA-seq data. We analyzed scRNA-seq data from mouse cortex cells to uncover subpopulation-specific responses to lipopolysaccharide treatment, and provide robust tools for multi-condition analysis within the muscat R package. INTRODUCTION A fundamental task in the analysis of single-cell RNA-sequencing (scRNA-seq) data is the identification of systematic transcriptional changes using differential expression analysis1. Such analyses are a critical step toward a deeper understanding of molecular responses that occur in development, after a perturbation, or in disease states2–5. Most current scRNA-seq differential expression methods are designed to test one set of cells against another (or more generally, multiple sets together), and can be used to compare cell subpopulations (e.g., for identifying marker genes) or across conditions (cells from one condition versus another)6. In such statistical models, the cells are the experimental units and thus represent the population that inferences will extrapolate to. Given the rise of multi-sample multi-group scRNA-seq datasets, where measurements are made on hundreds to thousands of cells per sample, the goal shifts to making sample-level inferences (i.e., experimental units are samples), in order to account for sample-to-sample as well as cell-to-cell variability and make conclusions that extrapolate to the samples rather than cells. We refer to this generally as differential state (DS) analysis, whereby a given subset of cells (termed hereafter as subpopulation) is followed across a set of samples (e.g., individuals) and experimental conditions (e.g., treatments), in order to identify subpopulation-specific responses, i.e., changes in cell state. DS analysis: (i) should be able to detect changes that only affect a single cell subpopulation, a subset of subpopulations or even a subset of cells within a single subpopulation; (ii) is intended to be an orthogonal analysis to clustering or cell subpopulation assignment; and, (iii) can be considered a separate analysis to the search for DA of subpopulations across conditions. We intentionally use the term subpopulation to be more generic than cell type7,8, which itself is meant to represent a discrete and stable molecular signature; however, the precise definition of cell type is widely debated2,3. In our framework, a subpopulation is simply a set of cells deemed to be similar enough to be considered as a group and where it is of interest to interrogate such sets of similarly defined cells across multiple samples and conditions. Therefore, cells from a scRNA-seq experiment are first organized into subpopulations, e.g., by integrating the multiple samples together9 and clustering or applying a subpopulation-level assignment algorithm10 or cell-level prediction11; clustering and manual annotation is also an option. Regardless of the model or the uncertainty in the subpopulation assignment, the discovery framework we describe provides a basis for biological interpretation and a path to discovering interesting expression patterns within subpopulations across samples. Even different subpopulation assignments of the same data could be readily interpretable. For example, T cells could be defined as a single (albeit diverse) cell subpopulation or could be divided into discrete subpopulations, if sufficient information to categorize the cells at this level of resolution is available. In either case, the framework presented here would focus on the subpopulation of interest and look for expression changes across conditions. This naturally introduces an interplay with the definition of cell types and states themselves (e.g., discrete states could be considered as types) and thus with the methods used to computationally or manually classify cells. Overall, our goal here is to explore the space of scRNA-seq datasets with several subpopulations and samples, in order to understand the fidelity of methods to discover cell state changes. It is worth noting that extensive workflows for DS analysis of high-dimensional cytometry data have been established12–15, along with a rich set of visualization tools and differential testing methods13,16–18, and applied to, for example, unravel subpopulation-specific responses to immunotherapy19. Notably, aggregation-based methods (e.g., representing each sample as the median signal from all cells of a given subpopulation) compare favorably in (cytometry) DS analysis to methods that run on full cell-level data17; however, in the cytometry case, only a limited range of cell-level and aggregation approaches were tested, only simplistic regimes of differential expression were investigated (e.g., shifts in means), and the number of features measured with scRNA-seq is considerably higher (with typically fewer cells). In scRNA-seq data, aggregating cell-level counts into sample-level “pseudobulk” counts for differential expression is not new; pseudobulk analysis has been applied to discover cell-type-specific responses of lupus patients to IFN-β stimulation20 and in mitigating plate effects by summing read counts in each plate21. In these cases, pseudobulk counts were used as input to bulk RNA-seq differential engines, such as edgeR22, DESeq223, or limma-voom24,25. Also, non-aggregation methods have been proposed, e.g., mixed models (MMs) were previously used on cell-level scRNA-seq expression data26 to separate sample and batch effects, and variations on such a MM could be readily applied for the sample-level inferences that are considered here. Various recent related developments have taken place: a compositional model was proposed to integrate cell type information into differential analysis, although replication was not considered27; a multivariate mixed-effects model was proposed to extend univariate testing regimes28; and, a tool called PopAlign was introduced to estimate low-dimensional mixtures and look for state shifts from the parameters of the mixture distributions29. Ultimately, there is scope for alternative methods to be applied to the discovery of interesting single-cell state changes. In existing comparison studies of scRNA-seq differential detection methods6,30,31, analyses were limited to comparing groups of cells and had not explicitly considered sample-level inferences or aggregation approaches. The rapid uptake of new single-cell technologies has driven the collection of scRNA-seq datasets across multiple samples. Thus, it remains to be tested whether existing methods designed for comparing expression in scRNA-seq data are adequate for such cross-sample comparisons, and in particular, how sensitive aggregation methods are to detect subpopulation-level responses. In this study, we developed a simulation framework, which is anchored to a reference dataset, that mimics various characteristics of scRNA-seq data and used it to evaluate 16 DS analysis methods (see Supplementary Table 1) across a wide range of simulation scenarios, such as varying the number of samples, the number of cells per subpopulation, and the magnitude and type of differential expression pattern introduced. We considered two conceptually distinct representations of the data for each subpopulation, cell-level or sample-level, and from these, made sample-level inferences. On cell-level data, we