Skip to content

Commit

Permalink
Facilitate iteration in differential subworkflow (#7115)
Browse files Browse the repository at this point in the history 
* illustrate differential iteration

* Combine matrix metas with contrast metas in differential modules

* Propose different iteration approach

* Flip input order

* Add multi-input test

* Update module snaps

* update meta.yml

* Update module meta.ymls

* See if different output prefix for limma_norm helps

* [skip ci] move fc to iterable channel

* update snapshots

* Fix metas

* update meta handling

* Put modules back to the way they were

* Try publishing logic fixes

* Try disabling R object save

* Revert "Try disabling R object save"

This reverts commit 71f804d.

* try fresh docker image URI

* Tidy up configs

* Don't repeat normalisation ops

* Add correct blocking var consideration to configs

* Add correct blocking var consideration to configs (snapshots)

* remove blank lines

* Corrections to channel structure, config

* Adding author/ maintainer block pending other author suggestions

* Add additional coauthors
  • Loading branch information
@pinin4fjords
pinin4fjords authored Dec 11, 2024
1 parent 7b32b09 commit 784b3a3
Show file tree
Hide file tree
Showing 9 changed files with 907 additions and 322 deletions.
2 changes: 1 addition & 1 deletion modules/nf-core/limma/differential/main.nf
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ process LIMMA_DIFFERENTIAL {
conda "${moduleDir}/environment.yml"
container "${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ?
'https://community-cr-prod.seqera.io/docker/registry/v2/blobs/sha256/af/afd9579a0ff62890ff451d82b360d85e82a0d61a3da40736ee0eee4e45926269/data' :
'community.wave.seqera.io/library/bioconductor-limma_bioconductor-edger:54dd09078d5db3b3' }"
'community.wave.seqera.io/library/bioconductor-edger_bioconductor-limma:176c202c82450990' }"

input:
tuple val(meta), val(contrast_variable), val(reference), val(target)
Expand Down
188 changes: 93 additions & 95 deletions subworkflows/nf-core/abundance_differential_filter/main.nf
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,113 +8,111 @@ include { DESEQ2_DIFFERENTIAL } from '../../../modules/nf-core/d
include { DESEQ2_DIFFERENTIAL as DESEQ2_NORM } from '../../../modules/nf-core/deseq2/differential/main'
include { CUSTOM_FILTERDIFFERENTIALTABLE } from '../../../modules/nf-core/custom/filterdifferentialtable/main'

// Combine meta maps, including merging non-identical values of shared keys (e.g. 'id')
def mergeMaps(meta, meta2){
(meta + meta2).collectEntries { k, v ->
meta[k] && meta[k] != v ? [k, "${meta[k]}_${v}"] : [k, v]
}
}

workflow ABUNDANCE_DIFFERENTIAL_FILTER {
take:
ch_abundance // [ meta_exp, counts ] with meta keys: method, args_diff
ch_transcript_lengths // [ meta_exp, transcript_lengths]
ch_control_features // [ meta_exp, control_features]
ch_samplesheet // [ meta_exp, samplesheet ]
ch_contrasts // [ meta_contrast, contrast_variable, reference, target ]
differential_method // limma, deseq2, propd
FC_threshold // float
padj_threshold // float

main:

// initialize empty results channels
ch_normalised_matrix = Channel.empty()
ch_variance_stabilised_matrix = Channel.empty()
ch_model = Channel.empty()

ch_results_genewise = Channel.empty()
ch_results_genewise_filtered = Channel.empty()
ch_versions = Channel.empty()

// Derive some commonly used derived channels
ch_samples_and_matrix = ch_samplesheet.join(ch_abundance).first()

logFC_column = null
padj_column = null

// ----------------------------------------------------
// Perform differential analysis with DESeq2
// ----------------------------------------------------

if (differential_method == 'deseq2') {

logFC_column = "log2FoldChange"
padj_column = "padj"

// Run the module once to generate a normalised matrix. We can't just
// use e.g. the first run of DESEQ_DIFFERENTIAL, because it may remove
// some samples

DESEQ2_NORM (
ch_contrasts.first(),
ch_samples_and_matrix,
ch_control_features,
ch_transcript_lengths
)
// Things we may need to iterate
ch_input // [[meta_input], counts, analysis method, fc_threshold, padj_threshold]

ch_normalised_matrix = DESEQ2_NORM.out.normalised_counts
ch_variance_stabilised_matrix = DESEQ2_NORM.out.rlog_counts.concat(DESEQ2_NORM.out.vst_counts)
// Workflow-wide things, we don't need to iterate
ch_samplesheet // [ meta_exp, samplesheet ]
ch_transcript_lengths // [ meta_exp, transcript_lengths]
ch_control_features // [meta_exp, control_features]
ch_contrasts // [[ meta_contrast, contrast_variable, reference, target ]]

// Run the DESeq differential module

DESEQ2_DIFFERENTIAL (
ch_contrasts,
ch_samples_and_matrix,
ch_control_features,
ch_transcript_lengths
)

ch_results_genewise = DESEQ2_DIFFERENTIAL.out.results
ch_model = DESEQ2_DIFFERENTIAL.out.model

ch_versions = ch_versions
.mix(DESEQ2_DIFFERENTIAL.out.versions)

} else if (differential_method == 'limma'){

logFC_column = "logFC"
padj_column = "adj.P.Val"

// Run the module once to generate a normalised matrix. We can't just
// use e.g. the first run of LIMMA_DIFFERENTIAL, because it may remove
// some samples
main:

LIMMA_NORM (
ch_contrasts.first(),
ch_samples_and_matrix
)
// Set up how the channels crossed below will be used to generate channels for processing
def criteria = multiMapCriteria { meta_input, abundance, analysis_method, fc_threshold, padj_threshold, meta_exp, samplesheet, meta_contrasts, variable, reference, target ->
samples_and_matrix:
meta_map = meta_input + [ 'method': analysis_method ]
[meta_map, samplesheet, abundance]
contrasts:
meta_map = mergeMaps(meta_contrasts, meta_input) + [ 'method': analysis_method ]
[ meta_map, variable, reference, target ]
filter_params:
meta_map = mergeMaps(meta_contrasts, meta_input) + [ 'method': analysis_method ]
[meta_map, [ 'fc_threshold': fc_threshold, 'padj_threshold': padj_threshold ]]
}

ch_normalised_matrix = LIMMA_NORM.out.normalised_counts
// For differential we need to cross the things we're iterating so we run
// differential analysis for every combination of matrix and contrast
inputs = ch_input
.combine(ch_samplesheet)
.combine(ch_contrasts)
.multiMap(criteria)

// We only need a normalised matrix from one contrast. The reason we don't
//just use the output from the first differential is that the methods can
// subset matrices
norm_inputs = ch_input
.combine(ch_samplesheet)
.combine(ch_contrasts.first()) // Just taking the first contrast
.multiMap(criteria)

// Perform normalization and differential analysis
DESEQ2_NORM(
norm_inputs.contrasts.filter{it[0].method == 'deseq2'}.first(),
norm_inputs.samples_and_matrix.filter{it[0].method == 'deseq2'},
ch_control_features.first(),
ch_transcript_lengths.first()
)

LIMMA_DIFFERENTIAL (
ch_contrasts,
ch_samples_and_matrix
)
ch_results_genewise = LIMMA_DIFFERENTIAL.out.results
ch_model = LIMMA_DIFFERENTIAL.out.model
LIMMA_NORM(
norm_inputs.contrasts.filter{it[0].method == 'limma'}.first(),
norm_inputs.samples_and_matrix.filter{it[0].method == 'limma'}
)

ch_versions = ch_versions
.mix(LIMMA_DIFFERENTIAL.out.versions)
DESEQ2_DIFFERENTIAL(
inputs.contrasts.filter{it[0].method == 'deseq2'},
inputs.samples_and_matrix.filter{it[0].method == 'deseq2'},
ch_control_features.first(),
ch_transcript_lengths.first()
)

}
LIMMA_DIFFERENTIAL(
inputs.contrasts.filter{it[0].method == 'limma'},
inputs.samples_and_matrix.filter { it[0].method == 'limma' }
)

// Combine results
ch_results = DESEQ2_DIFFERENTIAL.out.results.mix(LIMMA_DIFFERENTIAL.out.results)
ch_normalised_matrix = DESEQ2_NORM.out.normalised_counts.mix(LIMMA_NORM.out.normalised_counts)
ch_model = DESEQ2_DIFFERENTIAL.out.model.mix(LIMMA_DIFFERENTIAL.out.model)
ch_versions = DESEQ2_DIFFERENTIAL.out.versions
.mix(LIMMA_DIFFERENTIAL.out.versions)

// Extract the fc and pval filters from the metamap we stashed them in
ch_diff_filter_params = ch_results
.join(inputs.filter_params)
.multiMap { meta, results, filter_meta ->
filter_input: [meta + filter_meta, results]
fc_column: meta.method == 'deseq2' ? 'log2FoldChange' : 'logFC'
padj_column: meta.method == 'deseq2' ? 'padj' : 'adj.P.Val'
fc_threshold: filter_meta.fc_threshold
padj_threshold: filter_meta.padj_threshold
}

// Filter differential results
CUSTOM_FILTERDIFFERENTIALTABLE(
ch_results_genewise,
logFC_column,
FC_threshold,
padj_column,
padj_threshold
ch_diff_filter_params.filter_input,
ch_diff_filter_params.fc_column,
ch_diff_filter_params.fc_threshold,
ch_diff_filter_params.padj_column,
ch_diff_filter_params.padj_threshold
)

emit:
results_genewise = ch_results_genewise // channel: [ val(meta), path(results) ]
results_genewise_filtered = CUSTOM_FILTERDIFFERENTIALTABLE.out.filtered // channel: [ val(meta), path(filtered_results) ]
normalised_matrix = ch_normalised_matrix // channel: [ val(meta), path(normalised_matrix) ]
variance_stabilised_matrix = ch_variance_stabilised_matrix // channel: [ val(meta), path(variance_stabilised_matrix) ] (Optional)
model = ch_model // channel: [ val(meta), path(model) ]
versions = ch_versions // channel: [ path(versions.yml) ]
results_genewise = ch_results.map{meta, results -> [meta, results] }
results_genewise_filtered = CUSTOM_FILTERDIFFERENTIALTABLE.out.filtered
normalised_matrix = ch_normalised_matrix
variance_stabilised_matrix = DESEQ2_NORM.out.rlog_counts.mix(DESEQ2_NORM.out.vst_counts)
model = ch_model
versions = ch_versions
}
Loading

0 comments on commit 784b3a3

Please sign in to comment.