2
$\begingroup$

I’m trying to run a Nextflow workflow in a custom Docker container. Without Docker, the workflow succeeds. But running it inside the container leads to an error because a dependent file cannot be found.

Here’s a minimal script to illustrate the issue (similar to the basic Nextflow script example):

params.reference = "/data/reference/GRCh38.fa"

reference = file(params.reference)
fastq_channel = Channel.fromFilePairs('/data/fastq/*_R{1,2}.fastq.gz')

process map_reads {
    input:
    set sample, '?.fastq.gz' from fastq_channel

    output:
    file("${sample}.sam") into sam_channel

    script:
    """
    echo 'bwa mem -Y "$reference" 1.fastq.gz 2.fastq.gz >${sample}.sam'
    stat -t "${reference}.bwt"
    """
}

Running this inside a container will cause the path of the query, /some/fastq/, to be mounted inside the container. However, the path to the database, /data/reference, will not be mounted. Consequently, I get the following execution error:

stat: cannot stat 'GRCh38-chr21.fa.bwt': No such file or directory

Without the container, the command succeeds.


I understand why the error occurs: Nextflow doesn’t track the dependency on $reference inside the command, and thus doesn’t mount its path in the container.

My question is, how do I fix this? The documentation is silent on this regard, and none of the workflows I can find online address the problem. For example, the GATK4 IARC workflows would presumably have the same issue, and aren’t configured for use with Docker.

Just adding the reference as a process input is insufficient, since BWA requires various other index files. Am I supposed to add these all manually as inputs?

$\endgroup$
1
$\begingroup$

The following workaround seems to work but ideally I don’t want to have to change the script portion of my process so that it works with Docker.

params.ref = 'data/reference/GRCh38.fa'

ref = file(params.ref)
ref_index_dir = ref.parent
fastq_channel = Channel.fromFilePairs('data/fastq/*_R{1,2}.fastq.gz')

process map_reads {
    input:
    set sample, '?.fastq.gz' from fastq_channel
    file ref
    file ref_index_dir

    output:
    file("${sample}.sam") into sam_channel

    script:
    """
    echo 'bwa mem -Y "reference/$ref" 1.fastq.gz 2.fastq.gz >${sample}.sam'
    stat -t reference/${ref}.bwt && touch ${sample}.sam
    """
}

Here, reference inside the script section refers to the parent directory name of the reference, and needs to be adjusted accordingly. Having to refer to reference/$ref is annoying but at least it works — both inside a Docker container and outside of it.

$\endgroup$
1
$\begingroup$

Process input files need to be declared properly to allow nextflow to create the required docker mounts accordingly. Therefore the reference file should be declared in the process input block, eh:

input:
 set sample, file(fastq) from fastq_channel
 file reference 
$\endgroup$
  • $\begingroup$ The issue is that BWA for instance doesn’t just use the “reference” file, otherwise there’d be no issue. But it additionally derives various other files from the reference filename opaquely (i.e. these names are in principle implementation details and not known, except that they all have the reference filename as a prefix). $\endgroup$ – Konrad Rudolph Jun 10 at 11:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.