process SEQWISH_INDUCE { tag "$meta.id" label 'process_medium' conda (params.enable_conda ? 'bioconda::seqwish=0.7.6' : null) container "${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ? 'https://depot.galaxyproject.org/singularity/seqwish:0.7.6--h5b5514e_1' : 'quay.io/biocontainers/seqwish:0.7.6--h5b5514e_1' }" input: tuple val(meta), path(paf), path(fasta) output: tuple val(meta), path("*.gfa"), emit: gfa path "versions.yml" , emit: versions when: task.ext.when == null || task.ext.when script: def args = task.ext.args ?: '' def prefix = task.ext.prefix ?: "${meta.id}" def input = paf.join(',') // this ensures that we can actually input a // comma-separated list of PAF files as required by // https://github.com/nf-core/pangenome. If one wants to use this, // ensure that you put a ".collect()" behind your channel. // See https://github.com/nf-core/pangenome/blob/34149c6cdc19bce3a7b99f97c769d8986a8d429b/main.nf#L543 // for an example. """ seqwish \\ --threads $task.cpus \\ --paf-alns=$input \\ --seqs=$fasta \\ --gfa=${prefix}.gfa \\ $args cat <<-END_VERSIONS > versions.yml "${task.process}": seqwish: \$(echo \$(seqwish --version 2>&1) | cut -f 1 -d '-' | cut -f 2 -d 'v') END_VERSIONS """ }