featureCounts requires identical mate ids

featureCounts, a read-counting program, requires identical mate ids to identify a pair of read mates as correctly paired. However, FASTQ files generated from an SRA file with fastq-dump have different mate ids for each mate in a pair. The forward and reverse mate ids end with .1 and .2, respectively. I wrote a bash function to fix BAM files with this problem.

Problem

When you use the fastq-dump program to create FASTQ files from an SRA file:

fastq-dump -I --gzip --split-files file.sra > fastq-dump.log

The read identifiers have suffixes .1 and .2 for the forward and reverse strand reads, respectively. This BAM file has read identifiers with suffixes:

$ samtools view aligned.bam | head -n4 | cut -f1

SRR1032976.1.1
SRR1032976.1.2
SRR1032976.2.1
SRR1032976.2.2

featureCounts identifies paired mates when two reads have identical read identifiers. The suffixes make the identifiers unique, so featureCounts treats them as two separate fragments instead of one fragment.

It matters for the purpose of computing expression values, since a single fragment will sometimes be double-counted and sometimes not. If a gene has one read pair and one singleton, its count will be 3 instead of the correct 2 due to this suffix bug.

Solution

I wrote a bash function to fix the problem:

fix_mate_ids() {
  local original="$1"
  local fixed="${original}.fixed.bam"
  (
    samtools view -H "$original";
    samtools view "$original" \
      | awk 'BEGIN { FS=OFS="\t" } { sub(/[12]$/, "", $1) } 1'
  ) \
    | samtools view -Shb -@4 - > "$fixed" \
    && mv -f "$fixed" "$original"
}

Use it like this:

$ fix_mate_ids aligned.bam

[samopen] SAM header is present: 25 sequences.

Now the BAM is fixed:

$ samtools view aligned.bam | head -n4 | cut -f1

SRR1032976.1.
SRR1032976.1.
SRR1032976.2.
SRR1032976.2.

Notice that the read identifier changed from SRR1032976.1.1 to SRR1032976.1. without the trailing 1 at the end. Now, the mates have identical ids, so featureCounts will count them as one fragment instead of two.