Single-cell mapper (scMappR): using scRNA-seq to infer cell-type specificities of differentially expressed genes

Dustin Sokolowski, Mariela Faykoo-Martinez, Lauren Erdman, Huayun Hou, Cadia Chan, Helen Zhu, Melissa M. Holmes, Anna Goldenberg, Michael D. WilsonbioRxiv 2020 · DOI

TL;DR

This paper introduces scMappR, an R package that bridges the gap between bulk RNA-seq and single-cell RNA-seq (scRNA-seq) to pinpoint which cell types are responsible for observed gene expression changes. By integrating bulk RNA-seq data with scRNA-seq reference datasets, scMappR provides cell-type specificity scores for differentially expressed genes (DEGs), offering deeper biological insights than bulk analysis alone.

What Problem Does This Paper Address?

Bulk RNA sequencing (RNA-seq) is a powerful tool for identifying genes that change expression between conditions, but it analyzes mixed cell populations from tissues. This makes it challenging to determine which specific cell types within that tissue are driving these expression changes. While single-cell RNA sequencing (scRNA-seq) can resolve this at a single-cell level, it faces technical and cost limitations, hindering its widespread adoption. This paper addresses the need for a method that can leverage the vast amount of existing bulk RNA-seq data and readily available scRNA-seq reference datasets to infer cell-type specificity for DEGs identified from bulk samples.

Methods

The core of scMappR is its ability to integrate bulk RNA-seq data with scRNA-seq reference data. The process involves:

  1. scRNA-seq Data Processing: scMappR includes a pipeline to process raw scRNA-seq count data into a "signature matrix." This matrix represents the likelihood of a gene being expressed in each cell type. This pipeline uses standard tools like Seurat for normalization, clustering, and marker identification.
  2. Cell-Type Proportion Estimation: RNA-seq deconvolution methods (specifically DeconRNAseq) are used to estimate the proportion of each cell type within a bulk RNA-seq sample, using the signature matrix.
  3. Cell-Weighted Fold-Changes (cwFold-changes): scMappR calculates "cell-weighted fold-changes" by re-weighting the fold-changes of DEGs from bulk RNA-seq. This re-weighting incorporates both the cell-type specificity of a gene (derived from the signature matrix) and the estimated cell-type proportions. A key innovation here is an orthogonalization step to make cell-type expression and proportion estimates independent.
  4. Cell-Type Specific Pathway Analysis: The calculated cwFold-changes are then used to rerank DEGs, allowing for cell-type specific pathway enrichment analysis using tools like g:Profiler.

Datasets

The authors utilized several datasets for benchmarking and demonstrating scMappR:

  • Monaco et al., 2019 dataset: This dataset, containing bulk RNA-seq of peripheral blood mononuclear cells (PBMCs) and scRNA-seq of 29 sorted immune cell types, was used for benchmarking scMappR's ability to improve cell-type specificity.
  • Valle Duraes et al., 2020 dataset: This dataset, involving bulk RNA-seq of mouse kidneys during regeneration, was used to showcase scMappR's application in identifying cell-type specific DEGs in a complex tissue regeneration process.
  • Tabula Muris dataset: This publicly available scRNA-seq dataset of mouse tissues was preprocessed and stored by scMappR to serve as a reference signature matrix for various tissues.

Key Insights and Ideas

  • Bridging the Bulk-scRNA-seq Divide: scMappR effectively bridges the gap between bulk RNA-seq and scRNA-seq, allowing researchers to extract more granular, cell-type specific insights from readily available bulk data.
  • Improved Cell-Type Specificity: The introduction of "cell-weighted fold-changes" (cwFold-changes) significantly improves the correlation between bulk DEGs and empirically measured cell-type specific DEGs, as demonstrated in the PBMC dataset.
  • Uncovering Hidden Signals: scMappR can reveal cell-type specific DEGs that might be masked in traditional bulk differential analysis, particularly for genes from minority cell populations.
  • Versatile Application: The tool is designed to be flexible, allowing users to input their own scRNA-seq data or utilize pre-computed signature matrices, making it adaptable to various research scenarios.

Key Results

  • scMappR's cwFold-changes showed a statistically significant increase in cell-type specificity compared to bulk DEGs in the PBMC dataset (average rho increase of 0.0471).
  • In the mouse kidney regeneration study, scMappR successfully identified known cell types involved in the process, such as fibroblasts, smooth muscle, and endothelial cells, and also highlighted immune cell involvement.
  • The tool could identify cell-type specific roles for DEGs even in cell types representing less than 5% of the total kidney cell population.
  • The analysis demonstrated that scMappR is robust to the number of cell types selected and the number of samples used, functioning appropriately even with sample sizes representative of routine RNA-seq studies.

Limitations and Future Work

  • Limited by Initial DEG List: scMappR is dependent on the initial list of DEGs from bulk RNA-seq. Genes that are upregulated in one cell type and downregulated in another might not be identified as DEGs and thus won't be analyzed by scMappR.
  • RNA-seq Deconvolution Assumptions: scMappR relies on RNA-seq deconvolution, which has its own assumptions and limitations, such as requiring more samples than cell types and assuming cell types in the signature matrix constitute the entire bulk sample. Sensitivity to sample number, cell-type markers, and data processing are also noted limitations.
  • Future Work: The authors suggest that as single-cell genomic technologies evolve (e.g., ATAC-seq, DNA methylation), tools like scMappR that integrate bulk and single-cell data will become increasingly important for analyzing these diverse datasets. The package is designed to be easily incorporated into existing pipelines and accommodate new data types.

Summary updated

Abstract

Abstract RNA sequencing (RNA-seq) is widely used to identify differentially expressed genes (DEGs) and reveal biological mechanisms underlying complex biological processes. RNA-seq is often performed on heterogeneous samples and the resulting DEGs do not necessarily indicate the cell types where the differential expression occurred. While single-cell RNA-seq (scRNA-seq) methods solve this problem, technical and cost constraints currently limit its widespread use. Here we present single cell Mapper (scMappR), a method that assigns cell-type specificity scores to DEGs obtained from bulk RNA-seq by integrating cell-type expression data generated by scRNA-seq and existing deconvolution methods. After benchmarking scMappR using RNA-seq data obtained from sorted blood cells, we asked if scMappR could reveal known cell-type specific changes that occur during kidney regeneration. We found that scMappR appropriately assigned DEGs to cell-types involved in kidney regeneration, including a relatively small proportion of immune cells. While scMappR can work with any user supplied scRNA-seq data, we curated scRNA-seq expression matrices for ∼100 human and mouse tissues to facilitate its use with bulk RNA-seq data alone. Overall, scMappR is a user-friendly R package that complements traditional differential expression analysis available at CRAN. Highlights scMappR integrates scRNA-seq and bulk RNA-seq to re-calibrate bulk differentially expressed genes (DEGs). scMappR correctly identified immune-cell expressed DEGs from a bulk RNA-seq analysis of mouse kidney regeneration. scMappR is deployed as a user-friendly R package available at CRAN.
BibTeX
@misc{sokolowski2020single,
  author = {Dustin Sokolowski and Mariela Faykoo-Martinez and Lauren Erdman and Huayun Hou and Cadia Chan and Helen Zhu and Melissa M. Holmes and Anna Goldenberg and Michael D. Wilson},
  title = {{Single-cell mapper (scMappR): using scRNA-seq to infer cell-type specificities of differentially expressed genes}},
  year = {2020},
  doi = {10.1101/2020.08.24.265298},
  howpublished = {bioRxiv},
}
Figure