A collection of scripts for processing single-cell RNA-seq read data, including protocols that use single-cell library prep kits to sequence barcoded pools of thousands of cells. ##SCRBseqPoolGeneCount.py For some experiments, there is an interest in understanding expression variation among cell populations assigned to various conditions, but the amount of RNA obtained from those populations is too small to serve as input to standard bulk RNA-seq library preparation protocols. In such cases, and where there is less interest in cell-to-cell variation, one can use single-cell library preparation methods to build libraries from pools of cells. MIT's BioMicroCenter currently does this using a modified version of the SCRB-seq protocol of Soumillion et al. (http://biorxiv.org/content/early/2014/03/05/003236) to generate such heaviliy 3'-end biased single-end RNA-seq libraries. Fastq files generated by this protocol produce reads with headers that contain (a) an index (identifying a set of plate rows) and a well id within those rows that together correspond to a specific sequenced pool of cells, and (b) a UMI. Under the assumption that a UMI should only map to the same genmoic location once, GetGeneCountsFromThreePrimeWellOverloadMIT.py takes a binary compressed (bam) version of a sequence alignment generated by bwa, and a bed file of gene boundaries for the reference genome, and extracts mRNA counts per gene. It requires samtools and bedtools to be in the path, either through a loaded module or via modifying the PATH environmental variable. This script only works properly with bwa alignments, as it uses the X0:i:1 tag to identify uniquely mapped reads. Using this code with other aligners will require changes to the extract_unique_mapped function.
The first step is to create a bwa index for your referene genome:
bwa index <genome.fasta>
Next,The first step is to map single-end reads. Once you have created a bwa genome index for the reference genome, a SLURM job submission script to Harvard's Odyssey cluster with two cmd line arguments (path to indexed reference fasta and fastq read file) would look something like this:
#!/bin/bash
#SBATCH -N 1
#SBATCH -n 8
#SBATCH -t 04:00:00
#SBATCH -p serial_requeue
#SBATCH --mem=16000
#SBATCH -o bwa_%A.out
#SBATCH -e bwa_%.err
source new-modules.sh
module purge
module load bwa/0.7.9a-fasrc01
module load samtools/1.3.1-fasrc01
# $1 = full path to reference fasta
# $2 = fastq file
ref=$1
fastq=$2
prefix=`echo $fastq |sed 's/.fq.gz//g'`
bwa aln -t 8 -l 24 $ref $fastq > ${prefix}.sai
bwa samse $ref ${prefix}.sai $fastq > ${prefix}.sam
samtools view -Sbh ${prefix}.sam > ${prefix}.bam
rm ${prefix}.sam
# NOTE: the amount of time required for the job, as specified by the -t argument, will vary depending upon the size of the file and the reference genome
The next step is to use SCRBseqPoolGeneCount.py to extract counts of unique UMI-read mapping start postions per gene. This script filters bwa mappings using three criteria: (1) multi-mapping reads are discarded, (2) to avoid ambiguity, mappings that intersect with the boundaries of > 1 gene are discarded, and (3) because the library construction protocol is strand-specific, we require that read mappings be on the same strand as the gene from which it putatively originates. Because 3' UTRs are often incompletely annotated, we also provide an option to extend 3' gene boundaries by a fixed number of bases, so as to include mapppings that fall just outside of the 3' gene boundary.
-
-bam binary compressed (bam) version of the sam alignment file generated by bwa samse
-
-gbed bed file of gene boundaries, which can be created, for example, by extracting the chromosome,start position, end position, and gene id from a gtf or gff annotation file using your unix parser of choice (awk,sed, etc.)
-
-3ext number of bases to extend the 3' gene boundaries (default = 0)
-
-intout name of output bedfile representing the intersection between uniquely mapped reads and gene boundaries
-
-count name of output count table
-
-strand whether to enforce that alignments and genes be on the same strand; default is True, as the Soumillion et al. protocol is strand specific. But in principle, this script could be adapted for libraries that are not stranded.
USAGE: python SCRBseqPoolGeneCount.py -bam <alignment.bam> -gbed <geneboundaries.bed> -3ext <integer> -intout <intersect.bed> -count <counts.table>