systemPipeR 2.11.6
This section provides general description and how to use this single cell RNAseq (scRNAseq) workflow. In the actual analysis report, this section is usually removed.
This scRNAseq workflow template is based on the counted 10x scRNA data. It means
this workflow expect users have output from cell counting programs, like cell ranger
.
The steps in this workflow include:
This template is modified from Seurat Tutorial.
Users want to provide here background information about the design of their scRNAseq project.
This report describes the analysis of a scRNAseq project studying …
Typically, users want to specify here all information relevant for the analysis of their scRNAseq study. This includes detailed descriptions of files, experimental design, reference genome, gene annotations, etc.
To create a Workflow within systemPipeR
, we can start by defining an empty
container and checking the directory structure:
library(systemPipeR)
sal <- SPRproject()
sal
To load the entire workflow with default settings, run
sal <- importWF(sal, "SPscrna.Rmd")
cat(crayon::blue$bold("To use this workflow, following R packages are expected:\n"))
cat(c("'Seurat", "ggplot2", "ggpubr", "patchwork", "dplyr", "tibble",
"readr'\n"), sep = "', '")
### pre-end
appendStep(sal) <- LineWise(code = {
library(systemPipeR)
library(Seurat)
library(dplyr)
library(ggplot2)
library(ggpubr)
library(patchwork)
}, step_name = "load_packages")
In this example, the single cell data is preprocessed/filtered 10x data from a healthy donor. Samples taken from peripheral blood mononuclear cells (PBMCs), about 3000 cells.
Dataset can be downloaded with this link: https://s3-us-west-2.amazonaws.com/10x.files/samples/cell/pbmc3k/pbmc3k_filtered_gene_bc_matrices.tar.gz
If the link is not working, visit 10x website for updated links.
For your real data, please preprocess and put the dataset inside data
directory
appendStep(sal) <- LineWise(code = {
# unzip the data
untar("data/pbmc3k_filtered_gene_bc_matrices.tar.gz", exdir = "data")
# load data
pbmc.data <- Read10X(data.dir = "data/filtered_gene_bc_matrices/hg19/")
# Use dim to see the size of dataset, example data has
# 2700 cells x 32738 genes
dim(pbmc.data)
}, step_name = "load_data", dependency = "load_packages")
We can plot to see how many cells have good expressions.
appendStep(sal) <- LineWise(code = {
at_least_one <- apply(pbmc.data, 2, function(x) sum(x > 0))
count_p1 <- tibble::as_tibble(at_least_one) %>%
ggplot() + geom_histogram(aes(x = value), binwidth = floor(nrow(pbmc.data)/400),
fill = "#6b97c2", color = "white") + theme_pubr(16) +
scale_y_continuous(expand = c(0, 0)) + scale_x_continuous(expand = c(0,
0)) + labs(title = "Distribution of detected genes",
x = "Genes with at least one tag")
count_p2 <- tibble::as_tibble(BiocGenerics::colSums(pbmc.data)) %>%
ggplot() + geom_histogram(aes(x = value), bins = floor(ncol(pbmc.data)/50),
fill = "#6b97c2", color = "white") + theme_pubr(16) +
scale_y_continuous(expand = c(0, 0)) + scale_x_continuous(expand = c(0,
0)) + labs(title = "Expression sum per cell", x = "Sum expression")
png("results/count_plots.png", 1000, 700)
count_p1 + count_p2 + patchwork::plot_annotation(tag_levels = "A")
dev.off()
}, step_name = "count_plot", dependency = "load_data")
appendStep(sal) <- LineWise(code = {
sce <- CreateSeuratObject(counts = pbmc.data, project = "pbmc3k",
min.cells = 3, min.features = 200)
# calculate mitochondria gene ratio
sce[["percent.mt"]] <- PercentageFeatureSet(sce, pattern = "^MT-")
}, step_name = "create_seurat", dependency = "load_data")
Filters applied here are:
min.cells = 3
: for a single gene, it must be presented in at least 3 cells.min.features = 200
: for any cell, it must at least have 200 genes with readable gene counts.In your real data, you may want to adjust the filters for different projects.
Some indicators are important to show the quality of the cell batch, for example
appendStep(sal) <- LineWise(code = {
png("results/qc1.png", 700, 700)
VlnPlot(sce, features = c("nFeature_RNA", "nCount_RNA", "percent.mt"),
ncol = 3)
dev.off()
qc_p1 <- FeatureScatter(sce, feature1 = "nCount_RNA", feature2 = "percent.mt")
qc_p2 <- FeatureScatter(sce, feature1 = "nCount_RNA", feature2 = "nFeature_RNA")
png("results/qc2.png", 700, 450)
qc_p1 + qc_p2 + patchwork::plot_annotation(tag_levels = "A")
dev.off()
}, step_name = "qc_seurat", dependency = "create_seurat")
Extreme numbers, outlines are likely low quality sequencing/bad cells.
As shown on the figure above, A is the ratio of percent mt vs. N count. Normally, no matter how many counts you can find from a cell, the percent mt gene ratio should always be stable. For cells with extremely high mt ratios, they should be removed. Looking at x-axis, for a single cell, the N counts should also be inside a reasonable range. Too low of counts indicates the cell gel bead is empty during barcoding. Too high of counts indicates a single gel bead had more than one cell during barcoding. Either these cases are bad, and cells of these should be removed.
B is N genes per cell vs. N counts per cell. Normally, if as the N counts increases, the N genes also increases. All cells should be connected to approximately one line. If you see more than one line, there may be problem with the sequencing, or cells are not from the same batch.
appendStep(sal) <- LineWise(code = {
# Based on the QC plots, please change the settings
# accordingly
sce <- subset(sce, subset = nFeature_RNA > 200 & nFeature_RNA <
2500 & nCount_RNA < 10000 & percent.mt <= 5)
}, step_name = "filter_cells", dependency = "create_seurat")
For RNAseq data, usually log transformation and TPM is performed. For scRNA, the normalization is
\[ Normalized\ count\ for\ a\ gene\ in\ a\ cell\ =log1p(\frac{Feature\ counts}{Total\ counts\ in\ a\ cell})\\ where\ log1p\ is\ ln(n+1) \]
appendStep(sal) <- LineWise(code = {
# scale.factor = 10000 is a convenient number for
# plotting, so the normalized counts is ranged between
# 0.xx to 10.
sce <- NormalizeData(sce, normalization.method = "LogNormalize",
scale.factor = 10000)
# compare counts before and after
count_p_norm <- tibble::as_tibble(BiocGenerics::colSums(sce$RNA@data)) %>%
ggplot() + geom_histogram(aes(x = value), bins = floor(ncol(pbmc.data)/50),
fill = "#6b97c2", color = "white") + theme_pubr(16) +
scale_y_continuous(expand = c(0, 0)) + scale_x_continuous(expand = c(0,
0)) + labs(title = "Total expression after normalization",
x = "Sum expression")
png("results/normalize_count_compare.png", 1000, 700)
count_p2 + count_p_norm + patchwork::plot_annotation(tag_levels = "A")
dev.off()
}, step_name = "normalize", dependency = "filter_cells")
After the normalization, expression is close to a normal distribution.
Some genes are highly variable among cells. These genes are the key for the research. Here, some top-ranked variable genes are calculated and will be used for downstream analysis.
appendStep(sal) <- LineWise(code = {
# 2000 is default
sce <- FindVariableFeatures(sce, selection.method = "vst",
nfeatures = 2000)
# top 10 variable genes
top10_var <- head(VariableFeatures(sce), 10)
# plot the top 2000 variable genes and mark top 10
png("results/variable_genes.png", 700, 600)
VariableFeaturePlot(sce) %>%
LabelPoints(points = top10_var, repel = TRUE) + theme_pubr(16)
dev.off()
}, step_name = "find_var_genes", dependency = "normalize")
To make the visualization within a reasonable range, scaling is performed. This changes the expression data points to be centered at 0 with standard deviation of 1.
appendStep(sal) <- LineWise(code = {
sce <- ScaleData(sce, features = rownames(sce))
}, step_name = "scaling", dependency = "find_var_genes")
appendStep(sal) <- LineWise(code = {
# only use the top 2000 genes, first 50 PCs (default)
sce <- RunPCA(sce, features = VariableFeatures(object = sce),
npcs = 50)
# we can use following command to see first 5 genes in
# each PC
print(sce$pca, dims = 1:5, nfeatures = 5)
}, step_name = "pca", dependency = "scaling")
appendStep(sal) <- LineWise(code = {
# plot PCA overview
png("results/pca_overview.png", 500, 500)
DimPlot(sce, reduction = "pca")
dev.off()
# plot top contributed genes in PC 1 and 2
png("results/pca_loadings.png", 700, 550)
VizDimLoadings(sce, dims = 1:2, reduction = "pca")
dev.off()
# we can also use heatmap to show top genes in
# different PCs
png("results/pca_heatmap.png", 700, 700)
DimHeatmap(sce, dims = 1:6, cells = ncol(sce)/5, balanced = TRUE,
slot = "scale.data")
dev.off()
}, step_name = "pca_plots", dependency = "pca")
After PCA, the dimension has reduced from tens of thousands (features) to 50. Before, we do the clustering, we can further reduced the PCs to only a few important ones, so the computation work can be easier. How to choose the best number of PCs is hard to decide. Here, the JackStraw plot and Elbow plot can be good ways to help us make the decision.
appendStep(sal) <- LineWise(code = {
# for demo purposes, only a few replicates are used to
# speed up the calculation in your real data, use
# larger number like 100, etc.
sce <- JackStraw(sce, num.replicate = 30)
sce <- ScoreJackStraw(sce, dims = 1:20)
png("results/jackstraw.png", 660, 750)
JackStrawPlot(sce, dims = 1:20)
dev.off()
png("results/elbow.png", 500, 500)
ElbowPlot(sce)
dev.off()
}, step_name = "choose_pcs", dependency = "pca")
There is a gap of values between PCs 1-6 and other PCs. Therefore PCs 1-6 are the most important. One could also include PCs 7-13 since they have \(p\leq0.05\). One would need to adjust the PC choice based on different datasets.
The Elbow plot also shows there is turning point around PC 7-8. All other following PCs did not change too much.
It seems we should choose PC 6 or 7 as the cutoff point based on plots above. However, Seurat documents recommend to combine methods above with GSEA. Some cell populations may relate to genes in later PCs, e.g. dendritic cell and NK aficionados are linked to PCs 12, 13. Therefore, if your computational resources allow, it would be good to include a few more PCs.
The Seurat documents recommend to repeat with 10, 15 or PCs. However, in our experiences, the results did not change significantly.
In the downstream analysis, this template choose 10 PCs.
Seurat uses graph-based methods like KNN to find the clusters.
appendStep(sal) <- LineWise(code = {
sce <- FindNeighbors(sce, dims = 1:10)
# resolution 0.4-1.2 good for 3000 cells, if you have
# more cells, increase the number will give you more
# clusters
sce <- FindClusters(sce, resolution = 0.5)
}, step_name = "find_clusters", dependency = "pca")
appendStep(sal) <- LineWise(code = {
sce <- RunUMAP(sce, dims = 1:20)
png("results/")
p_umap <- DimPlot(sce, reduction = "umap", label = TRUE)
sce <- RunTSNE(sce, dims = 1:20)
p_tsne <- DimPlot(sce, reduction = "tsne", label = TRUE)
png("results/plot_clusters.png", 1000, 570)
p_umap + p_tsne
dev.off()
}, step_name = "plot_cluster", dependency = "find_clusters")
Find genes that represent different clusters.
appendStep(sal) <- LineWise(code = {
# find markers for every cluster compared to all
# remaining cells, report only the positive ones
sce.markers <- FindAllMarkers(sce, only.pos = TRUE, min.pct = 0.25,
logfc.threshold = 0.25)
sce.markers %>%
group_by(cluster) %>%
top_n(n = 2, wt = avg_log2FC)
}, step_name = "find_markers", dependency = "find_clusters")
appendStep(sal) <- LineWise(code = {
png("results/vlnplot.png", 600, 600)
VlnPlot(sce, features = c("MS4A1", "CD79A"))
dev.off()
png("results/marker_features.png", 700, 500)
FeaturePlot(sce, features = c("MS4A1", "GNLY", "CD3E", "CD14"))
dev.off()
# plot top 10 DEG genes in each cluster as a heatmap
top10_markers <- sce.markers %>%
group_by(cluster) %>%
top_n(n = 10, wt = avg_log2FC)
png("results/marker_heatmap.png", 1100, 700)
DoHeatmap(sce, features = top10_markers$gene) + NoLegend()
dev.off()
}, step_name = "plot_markers", dependency = "find_markers")
There are a few ways one can classify different clusters into different cell types.
The best way is you know what are the markers for targeting cell types.
CellMarker(http://biocc.hrbmu.edu.cn/CellMarker/) is a great source to find
markers of different cell types. If you do not know the markers, use singleR
package may be helpful. This package automatically classify clusters into cell
types. However, the accuracy is not promised.
For example, if we know the markers of different cell types as following:
Cluster | Markers | Cell Type |
---|---|---|
0 | IL7R, CCR7 | Naive CD4+ T |
1 | CD14, LYZ | CD14+ Mono |
2 | IL7R, S100A4 | Memory CD4+ |
3 | MS4A1 | B |
4 | CD8A | CD8+ T |
5 | FCGR3A, MS4A7 | FCGR3A+ Mono |
6 | GNLY, NKG7 | NK |
7 | FCER1A, CST3 | DC |
8 | PPBP | Platelet |
appendStep(sal) <- LineWise(code = {
new.cluster.ids <- c("Naive CD4 T", "CD14+ Mono", "Memory CD4 T",
"B", "CD8 T", "FCGR3A+ Mono", "NK", "DC", "Platelet")
names(new.cluster.ids) <- levels(sce)
sce <- RenameIdents(sce, new.cluster.ids)
png("results/marker_labels.png", 700, 700)
DimPlot(sce, reduction = "umap", label = TRUE, pt.size = 0.5) +
NoLegend()
dev.off()
}, step_name = "label_cell_type", dependency = c("plot_cluster",
"find_markers"))
appendStep(sal) <- LineWise(code = {
sessionInfo()
}, step_name = "wf_session", dependency = "label_cell_type")
To run the workflow, use runWF
function. It executes all the steps store in
the workflow container. The execution will be on a single machine without
submitting to a queuing system of a computer cluster.
sal <- runWF(sal, run_step = "mandatory") # remove `run_step` to run all steps to include optional steps
To check command-line tools used in this workflow, use listCmdTools
, and use listCmdModules
to check if you have a modular system.
The following code will print out tools required in your custom SPR project in the report. In case you are running the workflow for the first time and do not have a project yet, or you just want to browser this workflow, following code displays the tools required by default.
if (file.exists(file.path(".SPRproject", "SYSargsList.yml"))) {
local({
sal <- systemPipeR::SPRproject(resume = TRUE)
systemPipeR::listCmdTools(sal)
systemPipeR::listCmdModules(sal)
})
} else {
cat(crayon::blue$bold("Tools and modules required by this workflow are:\n"))
cat(c("NA"), sep = "\n")
}
## Tools and modules required by this workflow are:
## NA
This is the session information for rendering this report. To access the session information
of workflow running, check HTML report of renderLogs
.
sessionInfo()
## R version 4.4.1 (2024-06-14)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 22.04.4 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.20-bioc/R/lib/libRblas.so
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_GB LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: America/New_York
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats graphics grDevices utils datasets
## [6] methods base
##
## other attached packages:
## [1] BiocStyle_2.33.1
##
## loaded via a namespace (and not attached):
## [1] digest_0.6.36 R6_2.5.1
## [3] codetools_0.2-20 bookdown_0.40
## [5] fastmap_1.2.0 xfun_0.46
## [7] cachem_1.1.0 knitr_1.48
## [9] htmltools_0.5.8.1 rmarkdown_2.27
## [11] lifecycle_1.0.4 cli_3.6.3
## [13] sass_0.4.9 jquerylib_0.1.4
## [15] compiler_4.4.1 tools_4.4.1
## [17] evaluate_0.24.0 bslib_0.8.0
## [19] yaml_2.3.10 formatR_1.14
## [21] BiocManager_1.30.23 crayon_1.5.3
## [23] jsonlite_1.8.8 rlang_1.1.4