feat(annevo): add prediction, decoding, and scatter workflow

assets/catalog/catalog.md

@@ -109,6 +109,16 @@
 
 ---
 
+## Gene annotation
+
+- [annevo/annotation](../../modules/nextflow/annevo/annotation/main.nf) — One-step ANNEVO annotation (prediction + decoding). Prefer the split modules for production.
+- [annevo/prediction](../../modules/nextflow/annevo/prediction/main.nf) — ANNEVO nucleotide-level inference (`process_gpu`; CPU if no GPU profile).
+- [annevo/decoding](../../modules/nextflow/annevo/decoding/main.nf) — Decode ANNEVO H5 predictions into GFF3 on CPU.
+- [subworkflows/annevo](../../subworkflows/annevo/main.nf) — Scatter (none / chromosome / weighted), predict, decode, and gather GFF3.
+- [workflows/annevo](../../workflows/annevo/main.nf) — Standalone ANNEVO pipeline. Bundled ANNEVO is non-commercial.
+
+---
+
 ## Transcript assembly & quantification
 
 - [aletsch](../../modules/nextflow/aletsch/main.nf) — Assemble RNA-seq transcripts using Aletsch.

modules/nextflow/annevo/annotation/main.nf

@@ -0,0 +1,67 @@
+/*
+Copyright (c) 2026 The Hiller Lab at the Senckenberg Gessellschaft für Naturforschung
+Distributed under the terms of the Apache License, Version 2.0.
+*/
+
+/*
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ANNEVO_ANNOTATION — One-step ANNEVO annotation (prediction + decoding).
+
+    Useful for small genomes and for checking upstream behaviour. Prefer
+    ANNEVO_PREDICTION → ANNEVO_DECODING for scalable runs so the GPU is not
+    held during CPU decoding.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+*/
+
+process ANNEVO_ANNOTATION {
+    tag "$meta.id"
+    label 'process_high'
+    label 'process_gpu'
+
+    container "${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ?
+        '' :
+        'ghcr.io/hillerlab/annevo:latest' }"
+
+    input:
+    tuple val(meta), path(fasta)
+    val lineage
+
+    output:
+    tuple val(meta), path("*.gff3"), emit: gff
+    path "versions.yml",             emit: versions
+
+    when:
+    task.ext.when == null || task.ext.when
+
+    script:
+    def args    = [task.ext.args ?: '', meta.annevo_args ?: ''].findAll { it }.join(' ')
+    def prefix  = task.ext.prefix ?: "${meta.id}"
+    def model   = task.ext.model ?: (meta.model ?: "/opt/annevo/saved_model/ANNEVO_${lineage}.pt")
+    def workers = Math.max(1, (task.cpus as int) - 1)
+    """
+    annevo annotation \\
+        -g ${fasta} \\
+        -m ${model} \\
+        -l ${lineage} \\
+        -o ${prefix}.gff3 \\
+        -t ${task.cpus} \\
+        --num_workers ${workers} \\
+        ${args}
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+
+    stub:
+    def prefix = task.ext.prefix ?: "${meta.id}"
+    """
+    touch ${prefix}.gff3
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+}

modules/nextflow/annevo/decoding/main.nf

@@ -0,0 +1,60 @@
+/*
+Copyright (c) 2026 The Hiller Lab at the Senckenberg Gessellschaft für Naturforschung
+Distributed under the terms of the Apache License, Version 2.0.
+*/
+
+/*
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ANNEVO_DECODING — Convert ANNEVO nucleotide predictions into GFF3.
+
+    CPU-only. Keep this process separate from prediction so GPU allocations
+    can be released as soon as the H5 file is written.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+*/
+
+process ANNEVO_DECODING {
+    tag "$meta.id"
+    label 'process_high'
+
+    container "${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ?
+        '' :
+        'ghcr.io/hillerlab/annevo:latest' }"
+
+    input:
+    tuple val(meta), path(fasta), path(predictions)
+
+    output:
+    tuple val(meta), path("*.gff3"), emit: gff
+    path "versions.yml",             emit: versions
+
+    when:
+    task.ext.when == null || task.ext.when
+
+    script:
+    def args   = [task.ext.args ?: '', meta.annevo_decoding_args ?: ''].findAll { it }.join(' ')
+    def prefix = task.ext.prefix ?: "${meta.id}"
+    """
+    annevo decoding \\
+        -g ${fasta} \\
+        -p ${predictions} \\
+        -o ${prefix}.gff3 \\
+        -t ${task.cpus} \\
+        ${args}
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+
+    stub:
+    def prefix = task.ext.prefix ?: "${meta.id}"
+    """
+    touch ${prefix}.gff3
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+}

modules/nextflow/annevo/prediction/main.nf

@@ -0,0 +1,66 @@
+/*
+Copyright (c) 2026 The Hiller Lab at the Senckenberg Gessellschaft für Naturforschung
+Distributed under the terms of the Apache License, Version 2.0.
+*/
+
+/*
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+    ANNEVO_PREDICTION — Nucleotide-level ANNEVO inference.
+
+    Thin wrapper around `annevo prediction`. GPU is used when `-profile gpu`
+    exposes a device; otherwise PyTorch falls back to CPU. Does not insert
+    biological defaults such as `--overlap_pred`.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+*/
+
+process ANNEVO_PREDICTION {
+    tag "$meta.id"
+    label 'process_high'
+    label 'process_gpu'
+
+    container "${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ?
+        '' :
+        'ghcr.io/hillerlab/annevo:latest' }"
+
+    input:
+    tuple val(meta), path(fasta)
+    val lineage
+
+    output:
+    tuple val(meta), path("*.h5"), emit: predictions
+    path "versions.yml",           emit: versions
+
+    when:
+    task.ext.when == null || task.ext.when
+
+    script:
+    def args    = [task.ext.args ?: '', meta.annevo_args ?: ''].findAll { it }.join(' ')
+    def prefix  = task.ext.prefix ?: "${meta.id}"
+    def model   = task.ext.model ?: (meta.model ?: "/opt/annevo/saved_model/ANNEVO_${lineage}.pt")
+    def workers = Math.max(1, (task.cpus as int) - 1)
+    """
+    annevo prediction \\
+        -g ${fasta} \\
+        -m ${model} \\
+        -l ${lineage} \\
+        -p ${prefix}.h5 \\
+        --num_workers ${workers} \\
+        ${args}
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+
+    stub:
+    def prefix = task.ext.prefix ?: "${meta.id}"
+    """
+    touch ${prefix}.h5
+
+    cat <<-END_VERSIONS > versions.yml
+    "${task.process}":
+        annevo: \$( annevo --version | sed 's/annevo //g' )
+    END_VERSIONS
+    """
+}

modules/nextflow/fxsplit/main.nf

@@ -7,7 +7,8 @@ Distributed under the terms of the Apache License, Version 2.0.
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
     FXSPLIT — Split FASTX/FASTQ reads into chunks for parallel processing.
     Partitions read files into smaller chunks to enable parallel processing
-    across multiple CPUs.
+    across multiple CPUs. Set meta.headers or task.ext.headers to use
+    fxsplit -H (one output file per FASTA record).
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 */
 
@@ -35,15 +36,18 @@ process FXSPLIT {
     def args          = task.ext.args   ?: ''
     def prefix        = task.ext.prefix ?: "${meta.id}"
     def chunks        = task.ext.chunks ?: 400000
+    def header_mode   = (meta.headers == true) || (task.ext.headers ? true : false)
+    def mode_flag     = header_mode ? '-H' : "-c ${chunks}"
+    def suffix_flag   = header_mode ? '' : "--suffix ${prefix}"
     def gzip          = reads.name.endsWith('.gz') ? true : false
     """
     fxsplit \\
         $args \\
         -f $reads \\
-        -c $chunks \\
+        ${mode_flag} \\
         -t $task.cpus \\
         -C \\
-        --suffix ${prefix}
+        ${suffix_flag}
 
     if [ ${meta.singleton} == true ]; then
         for f in chunks/*fasta.gz; do
@@ -73,10 +77,14 @@ process FXSPLIT {
 
     if [ $gzip == true ]; then
         mkdir chunks/gz
-        mv chunks/*fast*.gz chunks/gz
+        mv chunks/*fast*.gz chunks/gz 2>/dev/null || mv chunks/*.gz chunks/gz
     else
         mkdir chunks/fx
-        mv chunks/*fast* chunks/fx
+        if [ "${header_mode}" = "true" ]; then
+            find chunks -maxdepth 1 -type f -exec mv {} chunks/fx/ \\;
+        else
+            mv chunks/*fast* chunks/fx
+        fi
     fi
 
     cat <<-END_VERSIONS > versions.yml