Demux – QSeq demultiplexer

Demux is a Hadoop progra to demultiplex data from multiplexed Illumina runs. Reads are demultiplexed by barcode and/or lane, according to the sample sheet configuration.

Features

• Barcode mismatches
• Separate data by read number (in addition to project/sample)
• Separate data for runs that aren’t multiplexed (only by lane)
• Input formats: qseq, fastq
• Output formats: qseq, fastq (independent of input, so you can transcode qseq to fastq or viceversa)
• Output compression: part files can be automatically compressed

Usage

To run Demux, the sample sheet needs to be accessible from all nodes. It’s easiest if you copy it to HDFS.

hadoop dfs -put sample_sheet.csv /user/me/

Then, run the seal demux command in the Seal distribution:

seal demux --sample-sheet /user/me/sample_sheet.csv /user/me/qseq_input /user/me/demuxed_output

The arguments are:

1. --sample_sheet: the HDFS path to the sample sheet file;
2. Input paths: files or directories, containing files for reads 1, 2 and 3 in the specified input format (default: qseq);
3. Output path: path to directory where demux will write its output. This directory must not already exist; demux will create it.

seal demux follows the normal Seal usage convention. See the section Program Usage for details.

Command line options

Short	Long	Description
-s	–sample-sheet <FILE>	Sample sheet for the experiment
-m	–mismatches <N>	Maximum number of acceptable barcode substitution errors (default: 0)
-ni	–no-index	Dataset doesn’t contain index reads. Sort reads only by lane (default: false)
-if	–input-format <FORMAT>	Input format name (qseq,fastq; default: qseq)
-of	–output-format <FORMAT>	Output format name(qseq,fastq; default: qseq)
-oc	–compress-output <CODEC>	Compress output files with CODEC (one of gzip bzip2, snappy, auto)
-sepr	–separate-reads	Generate separate directories for each read number (default: false)
-r	–num-reducers <INT>	Number of reduce tasks to use.
-sc	–seal-config <FILE> -D hbam.input.filter-failed-qc=true	Override default Seal config file Discard reads that failed the machine quality check.

Output

At the specified output path, seal demux creates a directory for each combination of project, sample and, optionally (see --separate-reads), read present in the input. If the sample sheet doesn’t contain a “Project” column (older CASAVA releases) the project name “DefaultProject” will be used.

Example:

DefaultProject/
  my_sample_1/
    read files
  my_sample_2
    read files

On the other hand, with --separate-reads:

DefaultProject/
  my_sample_1/
    1/
      read files
    2/
      read files
  my_sample_2
    1/
      read files
    2/
      read files

The project and sample directory names correspond to the sample and project specifed in the sample sheet for the read’s lane and barcode.

In addition, you may find an unknown directory containing qseq files with all the reads whose barcode was not found in the sample sheet (likely due to sequencing errors).

Inside the sample’s directory you’ll find regular qseq or fastq files, according to the --output-format option.

Note

demux can also compress its output files as it writes them; see --output-compression

Note

gzip files can be concatenated into a single larger archive. E.g.,

cat part-1.gz part-2.gz > whole.gz

Background

Multiplexed runs are used to sequence multiple samples together. Each sample’s fragments are tagged with a barcode sequence, which is annotated in a sample sheet. Illumina’s base calling software then emits 3 reads for each DNA/RNA fragment:

1. read 1
2. barcode
3. read 2

Illumina provides a utility to separate the multiplexed reads by barcode, but it is a simple serial implementation. Seal Demux replaces this utility with a Hadoop-based implementation which can drastically reduce run times when run on a cluster and can leverage HDFS storage like the rest of the Seal tools.

Demux loads the sample sheet and keeps track of which lane/barcode combination is associated with each sample. For each flow-cell location present in the input (tile / xpos / ypos), Demux gets the barcode (read 2), looks it up in the sample sheet and assigns the reads from that location to the appropriate project/sample.

Mismatches

Demux supports matching barcodes with substitution errors. You can specify the number of errors you’ll willing to tolerate with the --mismatches option. By default, no mismatches are tolerated.

The maximu number of mismatches that can be tolerated depends on the barcode sequences used in the run; the more “different” they are, the greater the number of mismatches that can be handled.

Sample Sheet

The sample sheet is a table in comma-separated value format with the following colums, in order:

FCID:flow-cell ID Lane:flow-cell lane SampleID:name of the sample SampleRef:sample reference Index:barcode used to tag this sample, excluding the last base, A Description:whatever you want Control:N Recipe:experimental protocol Operator:operator name Project:Project for the sample (optional. If absent, “DefaultProject” will be used.

Here’s an excerpt from a sample sheet file:

"FCID","Lane","SampleID","SampleRef","Index","Description","Control","Recipe","Operator"
"b02tgkkio",1,"csbb_001234","Human","ATCACG","Sequencing Project","N","tru-seq multiplex","Peter"
"b02tqacee",1,"csbb_004312","Human","CGATGT","Sequencing Project","N","tru-seq multiplex","Peter"

Counters

In addition to the counters from the Hadoop framework, Demux counts the number of reads found for each sample, and the unknowns. You’ll find them in the Sample reads counter group.

Configurable Properties

Demux does not have any program-specific configurable properties at the moment. You can still use its section to configure Hadoop property values specific to Demux.

Note

Config File Section Title: Demux