Sample pre-processing

Approximate time: 15 minutes

Learning Objectives

• Understand the steps taken to generate the Seurat object used as input for the workshop.

Sample data

For this workshop, we will be working with a single-cell RNA-seq dataset from Tseng et al, 2021. The data is available on GEO GSE160585. The files we will need to create the fully processed Seurat object include:

• Metadata csv file
• Counts matrix
• List of features (genes)
• List of cell barcodes

We have an entire workshop of materials that go through the whole process step-by-step on how to generate a similarly fully annotated, filtered dataset from single-cell RNA-seq data. In this lesson we provide mostly code, so you can reproduce the object yourself. If you want an in-depth explanation of each step we encourage you to peruse the materials linked above.

Pre-processing steps

1. Download and unzip the dataset from GEO using bash:

#!/bin/bash

# Create data directory to store downloaded files
mkdir -p data/filtered_counts

# Metadata csv file
wget wget https://ftp.ncbi.nlm.nih.gov/geo/series/GSE160nnn/GSE160585/suppl/GSE160585%5Fmetadata%5Ffor%5Fpseudotime%5Fand%5Fpseudobulk%5FDGE.csv.gz -O data/meta.csv.gz

# Features, barcodes, and counts matrix
wget https://ftp.ncbi.nlm.nih.gov/geo/series/GSE160nnn/GSE160585/suppl/GSE160585%5Ffiltered%5Fraw%5Fcounts%5Ffor%5Fpseudotime%5Fand%5Fpseudobulk%5FDGE.tar.gz -O data/filtered_counts.tar.gz

# Unzip and decompress the files
tar -xvf data/filtered_counts.tar.gz -C data/filtered_counts
gunzip data/meta.csv.gz

2. Data wrangling of the metadata

library(tidyverse)
meta <- read.csv("data/meta.csv", row.names=1)

# Celltype IDs have are formatted like: 
# {celltype}_{cluster}
# Removing the underscore
meta$celltype <- sub("_.*", "", meta$cluster_id)
meta <- select(meta, -c(cluster_id))

# The following columns in the metadata have duplicate values:
# - nCount_RNA = nUMI
# - nFeature_RNA = nGene
meta <- select(meta, -c(nUMI, nGene))

# Rename columns for more clarity
meta <- meta %>%
    rename(c("orig.ident"="sample", "sample"="condition"))


# Removing cluster resolutions that will not be used
cols <- c(
    "integrated_snn_res.0.1",
    "integrated_snn_res.0.4",
    "integrated_snn_res.0.6",
    "integrated_snn_res.0.8",
    "integrated_snn_res.1",
    # "integrated_snn_res.1.2",
    "integrated_snn_res.1.4",
    "integrated_snn_res.1.8",
    "seurat_clusters"
)

meta <- meta %>% select(-c(cols))

# Store clusters IDs as factors
meta$seurat_clusters <- meta$integrated_snn_res.1.2
sorted_cluster <- sort(as.integer(unique(meta$seurat_clusters)))
meta$seurat_clusters <- factor(meta$seurat_clusters, levels=sorted_cluster)

3. Generate Seurat object using downloaded files as input

library(Seurat)
library(Matrix)
set.seed(1454944673L) # Using the same seed used in the paper

# Load metadata, barcodes, features, and matrix into R
barcodes <- read.csv("data/filtered_counts/barcodes_filtered_raw_counts_for_pseudotime_and_pseudobulk_DGE.tsv", header=FALSE)
features <- read.csv("data/filtered_counts/genes_filtered_raw_counts_for_pseudotime_and_pseudobulk_DGE.tsv", header=FALSE)
counts <- readMM("data/filtered_counts/filtered_raw_counts_for_pseudotime_and_pseudobulk_DGE.mtx")

# Add gene and cell barcode information to count matrix
row.names(counts) <- features$V1
colnames(counts) <- barcodes$V1

# Create seurat object
seurat <- CreateSeuratObject(
    counts, 
    project = "GSE160585", 
    assay = "RNA", 
    meta.data = meta)

Note

In the next few steps we have provided the code to process the Seurat object. The parameters for each step were chosen based upon the descriptions provided in the Methods section of the paper.

4. Log-normalization and highly variable genes

# Log normalization
seurat <- NormalizeData(seurat)

# Identify the most variable genes
seurat <- FindVariableFeatures(seurat, 
                     selection.method = "vst",
                     nfeatures = 3000, 
                     verbose = FALSE)

5. SCTransform and regress out cell cycle scores

# Split seurat object by sample
split_seurat <- SplitObject(seurat, split.by = "condition")

# Allow R to use more memory
options(future.globals.maxSize = 4000 * 1024^2)

# Run SCTranform on each sample individually
for (i in 1:length(split_seurat)) {
    # Regress out cell cycle scores
    split_seurat[[i]] <- SCTransform(split_seurat[[i]], 
                                    vars.to.regress = c("S.Score", "G2M.Score"), 
                                    vst.flavor = "v2",
                                    variable.features.n = 3000)
}

6. CCA integration

# Select the most variable features to use for integration
integ_features <- SelectIntegrationFeatures(object.list = split_seurat, 
                                            nfeatures = 3000)
# Prepare the SCT list object for integration
split_seurat <- PrepSCTIntegration(object.list = split_seurat,
                                    anchor.features = integ_features)
# Find best buddies - can take a while to run
integ_anchors <- FindIntegrationAnchors(object.list = split_seurat,
                                        normalization.method = "SCT",
                                        anchor.features = integ_features)
# Integrate across conditions
seurat_integrated <- IntegrateData(anchorset = integ_anchors, 
                                    normalization.method = "SCT")

# Rejoin the layers in the RNA assay that we split earlier
seurat_integrated[["RNA"]] <- JoinLayers(seurat_integrated[["RNA"]])

7. PCA, nearest neighbors, UMAP

seurat_integrated <- RunPCA(seurat_integrated, verbose = FALSE)
seurat_integrated <- RunUMAP(seurat_integrated, dims = 1:50)
seurat_integrated <- FindNeighbors(seurat_integrated, dims = 1:50)

8. Save seurat object

Idents(seurat_integrated) <- "condition"
DefaultAssay(seurat_integrated) <- "RNA"

saveRDS(seurat_integrated, "data/BAT_GSE160585.rds")