vjcitn/pcmp

compare projection methods for scRNA-seq etc.

title	author	date	vignette	output
pcmp: projection comparison utilities for scRNA-seq	Vincent J. Carey, stvjc at channing.harvard.edu	`r format(Sys.time(), '%B %d, %Y')`	%\VignetteEngine{knitr::rmarkdown} %\VignetteIndexEntry{pcmp: projection comparison utilities for scRNA-seq} %\VignetteEncoding{UTF-8}	BiocStyle::html_document highlight number_sections theme toc pygments true united true

suppressPackageStartupMessages({
library(pcmp)
options(digits=3)
})

Introduction

Visualization of single-cell RNA-seq data is supported in many software environments and tools. pcmp was devised to address the following aims

• use Rstudio's crosstalk to foster linked/brushable selections in multiple panels
• record manual selections of clusters with indicator variables added in the colData of the input SingleCellExperiment instance
• provide statistical information on expression patterns that distinguish selected clusters, and on expression-based sources of variation within clusters
• for nuclear sequencing applications, support introduction and interpretation of measures of relative abundance of immature transcripts

The basic interface has the following appearance:

The input data are obtained from a SingleCellExperiment instance, in this case a stratified random sample of 1800 cells from the Allen Brain Atlas single-nucleus RNA-seq repositories from the ACC (anterior cingulate cortex) and VIS (primary visual cortex) brain regions.

The data are supplied in the pcmp package as the object sce300xx. The projections are retrieved from the reducedDims component of sce300xx, and are not computed by pcmp. Each point represents a single cell. Points are colored using palettes determined by crosstalk, and the groupings are determined by the stratby control, which gives access to all 'discrete' variables in the colData of the input SingleCellExperiment.

A couple of motivations for this app are visible in the following display, which was created by selecting a small group of mostly orange points in the center of dimensions 1,2 of the 2-dimensional t-SNE projection. You can see dashed box boundaries on the left panel. The color scheme is determined by the stratby selection. In this display color corresponds to tissue source: orange for primary visual cortex, blue for anterior cingulate cortex.

Cells are blurred in all projections if not selected.
Most of the orange cells in the t-SNE selection are clustered together in the UMAP projection on the right, but a few are scattered to other groups that seem more remote in the UMAP view.

Installation and execution

The package can be installed using R 3.5 or above with the command

devtools::install_github("vjcitn/pcmp")`

After successful installation, use

library(pcmp)
pcmpApp(sce300xx)

to work with the sample from the Allen Brain Atlas data.

In general, with any SingleCellExperiment instance with multiple projections represented in reducedDims, with equal numbers of dimensions per projection, and at least 4 such dimensions, pcmpApp should work. We do not know at present how well crosstalk scales with more than 2000 or so cells.

Additional features

Differential expression: selected cells vs. all others

A very simple approach for characterizing genes whose expression may distinguish a selected cluster is provided. We use limma on the log(assay(sce)+1) and keep the top 20 genes. The results for a given selection are only computed if the selTable tab is selected after the selection. You will have to wait until the table appears to make a new selection.

Here is the table that appears after the selTable tab is chosen after the selection shown in the previous display.

This table is stored in a special object in .GlobalEnv for later use.

Recording selections

Plotting selections after a session

pcmpApp returns an object of class SingleCellExperiment. The package includes a small serialized instance with some data from sce300xx, called vigAccum.

pcmp::vigAccum

We can plot our selections in a given projection using plotSelMap.

plotSelMap(pcmp::vigAccum, "TSNE2", 1, 2)

Biplots of discriminating expression patterns

Selections can be accompanied with lists of genes associated with cluster membership. By default these genes are developed using limma. The biplotSel function displays a PCA-based biplot using the expression data available for a selected subset of the discriminating genes for any cluster.

biplotSel(pcmp::vigAccum, selnum=4, ntop=10)