Mercurial > repos > fubar > egapx_runner
view nf/subworkflows/ncbi/only_gnomon.nf @ 2:a3b158471bd3 draft
planemo upload for repository https://github.com/ncbi/egapx commit 98875ef7eda9323fc9991970103954e9097d9e73
author | fubar |
---|---|
date | Sun, 04 Aug 2024 00:06:43 +0000 |
parents | d9c5c5b87fec |
children |
line wrap: on
line source
#!/usr/bin/env nextflow // gnomon-only nextflow script for EGAPx execution // route data to subworkflows nextflow.enable.dsl=2 include { setup_genome; setup_proteins } from './setup/main' include { get_hmm_params; run_get_hmm } from './default/get_hmm_params/main' include { chainer_wnode as chainer } from './gnomon/chainer_wnode/main' include { gnomon_wnode } from './gnomon/gnomon_wnode/main' include { prot_gnomon_prepare } from './gnomon/prot_gnomon_prepare/main' include { annot_builder } from './default/annot_builder/main' include { annotwriter } from './default/annotwriter/main' include { run_align_sort} from './default/align_sort_sa/main' params.intermediate = false workflow only_gnomon { take: genome // path to genome proteins // path to proteins, optional // Alternative groups of parameters, one of them should be set rnaseq_alignments // path to rnaseq_collapse'ed alignments protein_alignments // path to miniprot, filtered, sorted, alignments organelles // path to organelle list // Alternative parameters, one of them should be set // tax_id - NCBI tax id of the closest taxon to the genome // hmm_params - HMM parameters tax_id // NCBI tax id of the closest taxon to the genome hmm_params // HMM parameters hmm_taxid // NCBI tax id of the taxon of the HMM // softmask // softmask for GNOMON, optional task_params // task parameters for every task main: def (scaffolds, gencoll_asn, unpacked_genome, genome_asn) = setup_genome(genome, organelles, task_params.get('setup', [:])) // Protein alignments def unpacked_proteins def proteins_asn = [] if (proteins) { // miniprot plane (unpacked_proteins, proteins_asn) = setup_proteins(proteins, task_params.get('setup', [:])) } // Combine RNASeq and protein alignments def alignments if (protein_alignments && rnaseq_alignments) { print(rnaseq_alignments.getClass()) print(rnaseq_alignments) print(protein_alignments.getClass()) print(protein_alignments) alignments = Channel.of(rnaseq_alignments).combine(Channel.of(protein_alignments)) } else if (protein_alignments) { alignments = protein_alignments } else if (rnaseq_alignments) { alignments = rnaseq_alignments } else { print("error") } // GNOMON def effective_hmm if (hmm_params) { effective_hmm = hmm_params } else { tmp_hmm = run_get_hmm(tax_id) b = tmp_hmm | splitText( { it.split('\n') } ) | flatten c = b | last effective_hmm = c } chainer(alignments, effective_hmm, /* evidence_denylist */ [], /* gap_fill_allowlist */ [], scaffolds, /* trusted_genes */ [], genome_asn, proteins_asn, task_params.get('chainer', [:])) gnomon_wnode(scaffolds, chainer.out.chains, chainer.out.chains_slices, effective_hmm, [], softmask, genome_asn, proteins_asn, task_params.get('gnomon', [:])) def models = gnomon_wnode.out.outputs // prot_gnomon_prepare(models, task_params.get('prot_gnomon_prepare', [:])) // actual gnomon end but whatever annot_builder(gencoll_asn, models, genome_asn, task_params.get('annot_builder', [:])) def accept_asn = annot_builder.out.accept_asn annotwriter(accept_asn, [:]) annotwriter.out.annoted_file emit: out_files = annotwriter.out.annoted_file evidence = annot_builder.out.outputs }