fix(star): star_delete_fastq_after_alignment = false keeps the reads (v0.1.19)

assets/changelog/Changelog.md

@@ -61,6 +61,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ---
 
+## [0.1.19] - 2026-10-25
+
+### Fixed
+
+- `star_delete_fastq_after_alignment = false` had no effect. The STAR and ruSTAR subworkflows passed `params.star_delete_fastq_after_alignment ?: true`, and Groovy's `?:` turns `false` into `true`, so the second pass always deleted its input reads. The value is now passed as set.
+
+### Changed
+
+- With `star_delete_fastq_after_alignment = false`, the conversion just before the aligner also keeps its input, deacon's output: `BQTOOLS_DECODE` (STAR with CBQ reads) and `BQTOOLS_ENCODE_ALIGN` (`bqtools_encode_before_alignment`). Before, `-resume` after a failure could not reuse read preprocessing or alignment, because each step deletes its input, the output of an earlier task: resuming a failed 9-sample run (HLdesRot15A spleen) reran 19,153 of 19,177 tasks. With every keep switch on (usage, 4.6), a resumed `test-bqc-star` run reused all 61 tasks. The defaults still delete.
+- Usage: what each step deletes, and the switches `-resume` needs.
+- Version bumped to `0.1.19` in the pipeline manifest.
+
+---
+
 ## [0.1.18] - 2026-10-23
 
 ### Fixed

assets/docs/usage.md

@@ -259,10 +259,30 @@ greatly improves detection of novel splice sites.
 | `star_extra_align_args` | – | Any extra STAR arguments you want appended to the alignment command. |
 | `star_save_unaligned` | `false` | Reserved; declared but not consumed by any step. |
 | `star_keep_first_pass_bam` | `false` | Keep the BAM of the first alignment pass (extra disk; normally deleted). |
-| `star_delete_fastq_after_alignment` | `true` | Delete the aligned fastqs after the second pass to save disk. |
+| `star_delete_fastq_after_alignment` | `true` | Delete the aligner's input reads after the second pass to save disk, and the CBQ/FASTQ file they were converted from just before the aligner (`bqtools decode` for STAR, `bqtools_encode_before_alignment`). `false` keeps them. |
 | `star_twopass_junctions_file` | – | Path to a pre-curated junction file (STAR sjdb format, 4 columns). Providing one **skips the first pass** entirely and aligns once with your junctions. |
 | `star_make_coverage` | `true` | Produce coverage tracks (per-sample BigWig → merged BigWig in `05_coverage`). Disable to save time/disk if you do not need them. |
 
+**Disk or `-resume`.** To save disk, steps delete their inputs once their
+output is written: deacon deletes the trimmed reads, the conversion just before
+the aligner deletes deacon's output, the aligner deletes its input reads after
+the second pass, beaver deletes the per-sample GTFs, and the coverage steps
+delete the bedGraphs, the per-sample BigWigs and the merged `.wig`. These
+inputs are outputs of earlier tasks. After a failure, `nextflow run ... -resume`
+finds them gone and reruns those tasks and everything after them, i.e. from
+trimming on. To keep `-resume` working, keep them all:
+
+- `deacon_keep_fastp_fastq = true`
+- `star_delete_fastq_after_alignment = false`
+- `beaver_keep_aletsch_gtf = true`
+- `bigtools_keep_bedgraph = true`, `wigtools_keep_bigwig = true` and `wigtobigwig_keep_wig = true`
+- with `star_make_coverage = false`, also `aletsch_keep_bam = true`
+
+Every sample's trimmed and decontaminated reads then stay in the work directory
+until you remove it. Leaving out `bigtools_keep_bedgraph` is worse than a full
+rerun: the cached alignment has lost its bedGraph, so the resumed run silently
+skips coverage for that sample.
+
 ### 4.6a CBQ reads (bqtools + bqc)
 
 CBQ is the columnar BINSEQ read format. A paired sample collapses into a
@@ -291,7 +311,7 @@ in one `input_dir` is supported — each sample is routed by its own extension.
 | Parameter | Default | What it does |
 |-----------|---------|--------------|
 | `bqtools_encode_fastqs` | `false` | Encode FASTQ inputs to `.cbq` before QC. Your original files in `input_dir` are never deleted. |
-| `bqtools_encode_before_alignment` | `false` | Encode to `.cbq` after decontamination, just for the aligner (**ruSTAR only**). The intermediate deacon FASTQs **are** deleted (they are work-directory files, and nothing downstream removes them otherwise). |
+| `bqtools_encode_before_alignment` | `false` | Encode to `.cbq` after decontamination, just for the aligner (**ruSTAR only**). The intermediate deacon FASTQs **are** deleted (they are work-directory files, and nothing downstream removes them otherwise), unless `star_delete_fastq_after_alignment = false`. |
 | `bqc_min_length` | `15` | `--min-length`: reject reads shorter than this. Always sent — `bqc` refuses to run with no operation configured. |
 | `bqc_quality_tail` | `15` | `--quality-tail`: trim 3' bases until the trailing window reaches this Phred score. |
 | `bqc_poly_g` | `true` | `--poly-g`: trim G-rich 3' tails (two-colour chemistry). |

nextflow.config

@@ -41,7 +41,7 @@ manifest {
     mainScript      = 'main.nf'
     defaultBranch   = 'master'
     nextflowVersion = '!>=24.10.5'
-    version         = '0.1.18'
+    version         = '0.1.19'
 }
 
 validation {
@@ -158,7 +158,7 @@ params {
     star_extra_align_args                     = null       // extra arguments for STAR
     star_save_unaligned                       = false      // saves unaligned reads
     star_keep_first_pass_bam                  = false      // keep first pass bam
-    star_delete_fastq_after_alignment         = true       // delete fastq after second pass
+    star_delete_fastq_after_alignment         = true       // delete the aligner's input reads after the second pass [-resume needs false, see usage 4.6]
     star_twopass_junctions_file               = null       // path to pre-curated junction file [STAR sjdb 4-col] -> skips first pass
     star_make_coverage                        = true       // make coverage [bedgraph -> bigwig -> wig]
 

subworkflows/metassembly/main.nf

@@ -176,13 +176,14 @@ workflow METASSEMBLE {
                 .map { meta, reads, _ -> [meta, reads] }
 
         // Encode just before alignment: fastp and deacon still run on FASTQ and only the
-        // aligner input becomes CBQ. delete_input is true because the deacon FASTQs are
-        // work-dir intermediates that nothing downstream removes any more -- without it
-        // this path would raise peak disk instead of lowering it.
+        // aligner input becomes CBQ. The deacon FASTQs are work-dir intermediates that
+        // nothing downstream removes any more, so they are deleted (otherwise this path
+        // would raise peak disk instead of lowering it), unless the aligner keeps its input
+        // reads too (star_delete_fastq_after_alignment = false, for -resume).
         if (params.bqtools_encode_before_alignment) {
             BQTOOLS_ENCODE_ALIGN(
                 ch_processed_reads.processed_reads.filter { _meta, reads -> !isCbq(reads) },
-                true
+                params.star_delete_fastq_after_alignment
             )
 
             ch_versions = ch_versions.mix(BQTOOLS_ENCODE_ALIGN.out.versions.first())
@@ -195,11 +196,12 @@ workflow METASSEMBLE {
 
         // STAR cannot read CBQ. Decode after deacon so bqc + deacon stay on CBQ
         // and only the aligner input becomes gzipped FASTQ.
-        // delete_input is true because the deacon .cbq is a work-dir intermediate.
+        // The deacon .cbq is a work-dir intermediate: deleted unless the aligner keeps
+        // its input reads too (star_delete_fastq_after_alignment = false, for -resume).
         if (params.aligner == 'STAR') {
             BQTOOLS_DECODE(
                 ch_aligner_reads.filter { _meta, reads -> isCbq(reads) },
-                true
+                params.star_delete_fastq_after_alignment
             )
 
             ch_versions = ch_versions.mix(BQTOOLS_DECODE.out.versions)

subworkflows/rustar_alignment/main.nf

@@ -108,7 +108,7 @@ workflow RUSTAR_ALIGNMENT {
                 params.star_seq_library ?: '',
                 params.star_machine_type ?: '',
                 true, // keep bam -> we need it for Aletsch
-                params.star_delete_fastq_after_alignment ?: true
+                params.star_delete_fastq_after_alignment
             )
         } else {
             ch_rustar_second_pass_out = RUSTAR_ALIGN_2PASS(
@@ -121,7 +121,7 @@ workflow RUSTAR_ALIGNMENT {
                 params.star_seq_library ?: '',
                 params.star_machine_type ?: '',
                 true, // keep bam -> we need it for Aletsch
-                params.star_delete_fastq_after_alignment ?: true
+                params.star_delete_fastq_after_alignment
             )
         }
 

subworkflows/star_alignment/main.nf

@@ -107,7 +107,7 @@ workflow STAR_ALIGNMENT {
                 params.star_seq_library ?: '',
                 params.star_machine_type ?: '',
                 true, // keep bam -> we need it for Aletsch
-                params.star_delete_fastq_after_alignment ?: true
+                params.star_delete_fastq_after_alignment
             )
         } else {
             ch_star_second_pass_out = STAR_ALIGN_2PASS(
@@ -120,7 +120,7 @@ workflow STAR_ALIGNMENT {
                 params.star_seq_library ?: '',
                 params.star_machine_type ?: '',
                 true, // keep bam -> we need it for Aletsch
-                params.star_delete_fastq_after_alignment ?: true
+                params.star_delete_fastq_after_alignment
             )
         }