scRNA-seq

Raw data processing

cellranger : 10x Genomics’s Chromium system to process single cell RNA-seq raw data. Easy to use but need large computing resources and Linux environment.

Input : Fastq files, cellranger genome index

cellranger count/aggr

# cellranger count for transcriptome data processing 
cellranger count --id=SampleA_GEX \
--transcriptome=/path/to/reference/transcriptomes/Mouse_GEX_2020/refdata-gex-mm10-2020-A \
--fastqs=/path/to/raw_data/SampleA \
--sample=SampleA \
--expect-cells=10000

# cellranger aggr to aggregate multiple cellranger count output to one file 
cellranger aggr --id=Aggregate_GEX \
--csv=/path/to/aggregate_info/Aggregate_GEX.csv

Aggregate_GEX.csv example

library_id,molecule_h5
SampleA,/path/to/SampleA/outs/molecule_info.h5
SampleB,/path/to/SampleB/outs/molecule_info.h5
SampleC,/path/to/SampleC/outs/molecule_info.h5

Output

Aggregated Feature-Barcode Matrices

matrix.mtx
genes.tsv
barcodes.tsv

Read Cellranger output

library(Seurat)
obj.raw = Read10X('raw_data/Path/filtered_feature_bc_matrix/')
obj.srt = CreateSeuratObject(counts = obj.raw, project = 'project_name')

Doublet removal (by scrublet)

scrublet runs in python env.
Input file is the count matrix from seurat object.

# If needed, install python pacakges first

# Create and activate a new conda environment
conda create -n bioinfo_env python=3.8 -y
conda activate bioinfo_env

# Install pip if not already installed
conda install pip -y

# Install necessary packages with pip
pip install numpy pandas scanpy scrublet

# run python
python

import numpy as np
import pandas as pd
import scanpy as sc
import scrublet as scr

def run_scrublet(input_file, output_file, expected_doublet_rate=0.1):
    df = pd.read_csv(input_file, header=0, index_col=0)
    adata = sc.AnnData(df)
    
    # Set the expected_doublet_rate parameter
    sc.external.pp.scrublet(adata, expected_doublet_rate=expected_doublet_rate)
    
    # Save the observation (results) to CSV
    adata.obs.to_csv(output_file)

input_file = 'input.count.csv'
output_file = 'output.scr.csv'

# Adjust the expected_doublet_rate parameter
expected_doublet_rate = 0.1
run_scrublet(input_file, output_file, expected_doublet_rate)

Add the scrublet output information to the seurat object

## import scrublet result
df= read.csv('path/output.scr.csv', row.names = 1)

# If the previous process changes the name of cell ID, 
# Use the following code to correct it. 
rownames(df) = gsub(pattern = '_', replacement = '-' ,rownames(df))

## add doublet info to the srt obj

obj.srt[['doublet_score']] = df[rownames(obj.srt@meta.data),]$doublet_score
obj.srt[['predicted_doublet']] =df[rownames(obj.srt@meta.data),]$predicted_doublet

obj.srt %>% saveRDS("path/saved.obj.rds")

# You might want to check the number of doublets in the data
obj.srt@meta.data %>% select(predicted_doublet) %>% table()

Filtering by UMI & mitocondrial content ratio

QC Example