Tool: repaq: compress Illumina FASTQ to be as small as 1/4 of the gzip format.
gravatar for chen
2.4 years ago by
chen2.1k wrote:

repaq is an open source tool for FASTQ compression:

It repacks Illumina format FASTQ to a smaller binary file (.rfq), which can be further compressed by xz or pxz (.rfq.xz).

For NovaSeq data, the .rfq file can be much smaller than .fq.gz, and the compressing time is usually less than 1/5 of gzip compression.

The biggest advantage is that the .rfq file can be further compressed with xz, which is based on LZMA algorithm. The .rfq.xz file can be as small as 5% of the original FASTQ file, or smaller than 30% of the .fq.gz file. Note that usually the gz files are not compressible by xz.

This tool also supports non-Illumina format FASTQ (i.e. the BGI-SEQ format), but the compression ratio is not as good Illumina format FASTQ.

WARNING: be careful about using repaq for production before v1.0 is released, since its spec v1.0 has not been frozen.

take a look of the compression ratio

Here we demonstrate the compression ratio of two paired-end NovaSeq data. You can download these files and test locally.

See? The size of final nova.rfq.xz is only 3.39% of the original FASTQ files! You can decompress it and check the md5 to see whether they are identical!

Typically with one single CPU core, it takes less than 1 minute to convert nova.R1.fq + nova.R2.fq to nova.rfq, and takes less than 5 minutes to compress the nova.rfq to nova.rfq.xz by xz.

get repaq

download binary

This binary is only for Linux systems:

# this binary was compiled on CentOS, and tested on CentOS/Ubuntu
chmod a+x ./repaq

or compile from source

# get source (you can also use browser to download from master or releases)
git clone

# build
cd repaq

# Install
sudo make install


For single-end mode:

# compress
repaq -c -i in.fq -o out.rfq

# decompress
repaq -d -i in.rfq -o out.fq

For paired-end mode:

# compress
repaq -c -i in.R1.fq -I in.R2.fq -o out.rfq

# decompress
repaq -d -i in.rfq -o out.R1.fq -O out.R2.fq


  • -i and -I always denote the first and second input files, while -o and -O always denote the first and second output files.
  • the FASTQ input/output files can be gzipped if their names are ended with .gz.

for paired-end data. the .rfq file created in paired-end mode is usually much smaller than the sum of the .rfq files created in single-end mode for R1 and R2 respectively. To obtain high compression rate, please always use PE mode for PE data.

compress .rfq to .rfq.xz with xz

To get highest compression ratio (need at least 16G RAM):

xz --lzma2="dict=1000000000" in.rfq

To get normal ratio (need at least 1G RAM):

xz -9 in.rfq

The latest version of xz supports multithreading, so you can specify the thread number with -T option:

xz -T4 -9 in.rfq

You can also use pxz for parallel xz compressing:

pxz -9 in.rfq


  • lower compression ratio than -9 is not recommended, since it will not be faster. The difference is the RAM requirement.


repaq can read the input from STDIN, and write the output to STDOUT.

  • specify --stdin if you want to read the STDIN for compression or decompression.
  • specify --stdout if you want to output to the STDOUT for compression or decompression.
  • in decompression mode, if --stdout is specified, the output will be interleaved PE stream.
  • if the STDIN is an interleaved paired-end stream, specify --interleaved_in to indicate that.

Here gives you an example of compressing the interleaved PE output from fastp by directly using pipes:

fastp -i R1.fq -I R2.fq --stdout | repaq -c --interleaved_in --stdin --stdout | xz -z -c > out.rfq.xz

FASTQ Format compatibility

repaq was initially designed for compressing Illumina data, but it also works with data from other platforms, like BGI-Seq. To work with repaq, the FASTQ format should meet following condidtions:

  • only has bases A/T/C/G/N.
  • each FASTQ record has, and only has four lines (name, sequence, strand, quality).
  • the name and strand line cannot be longer than 255 bytes.
  • the number of different quality characters cannot be more than 127.

repaq works best for Illumina data directly output by bcl2fastq.

all options

  -i, --in1                    input file name (string [=])
  -o, --out1                   output file name (string [=])
  -I, --in2                    read2 input file name when encoding paired-end FASTQ files (string [=])
  -O, --out2                   read2 output file name when decoding to paired-end FASTQ files (string [=])
  -c, --compress               compress input to output
  -d, --decompress             decompress input to output
  -k, --chunk                  the chunk size (kilo bases) for encoding, default 1000=1000kb. (int [=1000])
      --stdin                  input from STDIN. If the STDIN is interleaved paired-end FASTQ, please also add --interleaved_in.
      --stdout                 write to STDOUT. When decompressing PE data, this option will result in interleaved FASTQ output for paired-end input. Disabled by defaut.
      --interleaved_in         indicate that <in1> is an interleaved paired-end FASTQ which contains both read1 and read2. Disabled by defaut.

# following options are used to check the consistency of the compressed data
  -p, --compare                compare the files read by read to check the compression consistency. <rfq_to_compare> should be specified in this mode.
  -r, --rfq_to_compare         the RFQ file to be compared with the input. This option is only used in compare mode. (string [=])
  -j, --json_compare_result    the file to store the comparison result. This is optional since the result is also printed on STDOUT. (string [=])

  -?, --help                   print this message
tool fastq repaq compress • 1.9k views
ADD COMMENTlink modified 2.4 years ago • written 2.4 years ago by chen2.1k

As with most compression methods, it's usually a case of compute time, vs space saved. Do you have any benchmarks against time for typical fastq files? - it would be useful by technology, RNAseq, WES, WGS, etc.

ADD REPLYlink written 2.4 years ago by andrew.j.skelton736.1k

Thanks, I will do the benchmark and post the results.

ADD REPLYlink written 2.4 years ago by chen2.1k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1589 users visited in the last hour