deinterleave FASTQ files

deinterleave_fastq.sh

#!/bin/bash
# Usage: deinterleave_fastq.sh < interleaved.fastq f.fastq r.fastq [compress]
# 
# Deinterleaves a FASTQ file of paired reads into two FASTQ
# files specified on the command line. Optionally GZip compresses the output
# FASTQ files using pigz if the 3rd command line argument is the word "compress"
# 
# Can deinterleave 100 million paired reads (200 million total
# reads; a 43Gbyte file), in memory (/dev/shm), in 4m15s (255s)
# 
# Latest code: https://gist.github.com/3521724
# Also see my interleaving script: https://gist.github.com/4544979
# 
# Inspired by Torsten Seemann's blog post:
# http://thegenomefactory.blogspot.com.au/2012/05/cool-use-of-unix-paste-with-ngs.html

# Set up some defaults
GZIP_OUTPUT=0
PIGZ_COMPRESSION_THREADS=10

# If the third argument is the word "compress" then we'll compress the output using pigz
if [[ $3 == "compress" ]]; then
  GZIP_OUTPUT=1
fi

if [[ ${GZIP_OUTPUT} == 0 ]]; then
  paste - - - - - - - -  | tee >(cut -f 1-4 | tr "\t" "\n" > $1) | cut -f 5-8 | tr "\t" "\n" > $2
else
  paste - - - - - - - -  | tee >(cut -f 1-4 | tr "\t" "\n" | pigz --best --processes ${PIGZ_COMPRESSION_THREADS} > $1) | cut -f 5-8 | tr "\t" "\n" | pigz --best --processes ${PIGZ_COMPRESSION_THREADS} > $2
fi

This is the fastest FASTQ deinterleaver I've seen as it uses native Linux commands paste, tee, tr and cut to process the file and there are no calculations required - just reformatting.

It assumes each read occupies 4 lines and read pairs are interleaved i.e. a block of 8 lines contain a pair of reads.

The bottleneck is usually disk IO so I've taken to mounting a tmpfs on my large memory machine and doing deinterleaving from there.

Using this script it takes:
58s to deinterleave a 4.1GByte FASTQ file residing on a tmpfs and writing to the same tmpfs.

Using this script it takes:
2m26s to deinterleave the same 4.1GByte FASTQ file residing on a RAID5 filesystem (5*15k rpm disks) and writing to the same filesystem.

Nice, indeed fast & simple!

I have thought about setting up and using named pipes (fifo's) to do deinterleaving on-the-fly so as to avoid the overhead of storing both interleaved and deinterleaved files. However I can't seem to get them to work properly! I've been trying something like this:

mkfifo f.fastq
mkfifo r.fastq

# deinterleave a paired-end/matepair FASTQ file to the named pipes
deinterleave_paste.sh < infile.fastq f.fastq r.fastq

# setup background process to read/process reads from each of the named pipes
cat r.fastq &
cat f.fastq &

Wow, I just wanted to say thank you for this simple yet fast and useful solution, thanks!

Indeed very nice!. FYI put the deinterleaver into background to make it work with fifos:

mkfifo f.fastq
mkfifo r.fastq

# deinterleave a paired-end/matepair FASTQ file to the named pipes (IN BACKGROUND)
deinterleave_paste.sh f.fastq r.fastq < infile.fastq &

# setup process to read/process reads from each of the named pipes (IN FOREGROUND)
cat r.fastq f.fastq

That's pretty cool

How to modify this to deinterleave a zipped fastq file (i.e. fastq.gz)?

this should work: gzip -dc test.fq.gz | deinterleave_fastq.sh f.fastq r.fastq

Or in python:

r1 = open(r1.fastq,'w')
r2 = open(r2.fastq,'w')
[r1.write(line) if (i % 8 < 4) else r2.write(line) for i, line in enumerate(open('interleaved.fastq'))]
fastq_1.close()
fastq_2.close()

I love this and I'd like to include it as part of my routine workflow. Could you please include a license or a comment line to indicate that this is OK with you (or what you're OK with)? Thanks!!

Awesome. I'm also using this. Thanks a lot!

I am having trouble getting the files to compress. I am running as:

gzip -dc test.fq.gz | deinterleave_fastq.sh f.fastq r.fastq compress

But my files are deleaving as fastq not fastq.gz. Should i be running as:

gzip -dc test.fq.gz | deinterleave_fastq.sh f.fastq.gz r.fastq.gz compress

This is really handy, I'd make a minimal improvement to protect against the input process being terminated at the end of a valid record (by chance). Just the addition of the following near the start:

set -e
set -o pipefail

If overly cautious you could add a bit more tee and confirm the lines written to read_1 and read_2 are equal number (if not it's a good indicator that the file may not be well formed) and the sum is that of the input:

# tmpdir would be necessary for the count_* files
tee >(paste - - - - - - - -  | tee >(cut -f 1-4 | tr "\t" "\n" | tee >(wc -l > count_a)\
 | gzip -c > $1) | cut -f 5-8 | tr "\t" "\n" | tee >(wc -l > count_b)\
 | gzip -c > $2) | wc -l > count_all
# then some checking of  values

Worked well and fast, but created 4 empty lines at the end of the de-interleaved files that interferred with downstream applications and had to be removed manually. Maybe something that can be fixed in a future version.
Took me a while to figure out the problem (error message of downstream application was cryptic), thus I though this might be useful for others to know.

Exactly like a-kroh mentioned, somehow it generates empty lines at the end of each output file. When I run the Trimmomatic, it exits during the run. Since I saw the comments, it does not take me too long to figure out what happened. This deinterleave_fastq.sh is pretty handy, it would be great if this problem got fixed.
I used
$ egrep -v '^$' EMPTYLINE.fastq > NO_EMPTYLINE.fastq
to remove the empty lines.

Pretty cool!
Thanks

Thank you all.
I've edited it a bit:

cat input.fastq | paste - - - - - - - - | tee | cut -f 1-4 | tr "\t" "\n" | egrep -v '^$' > R1.fastq
cat input.fastq | paste - - - - - - - - | tee | cut -f 5-8 | tr "\t" "\n" | egrep -v '^$' > R2.fastq