SVcaller

A production-ready Nextflow DSL2 pipeline for calling structural variants (SVs), copy number variants (CNVs), and SMN1/SMN2 copy numbers from Illumina paired-end WGS data (PE150, ≥30×).

Accepts FASTQ or pre-aligned BAM input. Produces a per-sample HTML report with an embedded Circos plot, annotated SV/CNV tables, SMN1/SMN2 classification, and optional GIAB benchmark metrics.

Self-contained per-sample report (SMAPB, an anonymised standardized reference sample): alignment QC, genome-wide Circos plot, structural-variant and copy-number summaries, and SMN1/SMN2 classification (here a homozygous SMN1 deletion consistent with SMA). Bootstrap CSS is inlined for air-gapped clinical distribution.

Features

Ensemble SV calling — 5-caller ensemble: Manta + DELLY + GRIDSS (core, all three must succeed) plus Scramble and MELT, each added only when it has calls. Merged with JASMINE. SvABA is not part of the ensemble — it is skipped by default and its VCF never reaches the merge; GRIDSS BND pairs auto-converted to typed DEL/DUP/INV
STR genotyping — ExpansionHunter (32 disease loci) + STRling genome-wide scanning
Dual CNV calling — CNVpytor + GATK gCNV with consensus merging
SMN1/SMN2 copy number — SMNCopyNumberCaller with 2+0 haplotype detection
Copy-number / blood-group traits — targeted normalized read depth for Rh factor (RHD), AMY1, GSTM1/GSTT1 null, and LPA KIV-2, emitted as stable per-sample contract files
Alpha-globin (HBA1/HBA2) — M8 measures alpha-thalassemia via per-segment depth, alpha-cluster junction search, and a pinned pathogenic-site pileup; emitted as a per-sample contract and a measurement-only report card (dosage 0–4, deletion reported as a group when depth cannot separate alleles, panel-named site hits). No clinical interpretation
Clinical annotation — AnnotSV 3.4.6 with SV PON (GIAB 7-sample), gnomAD-SV AF, SegDup, and ENCODE blacklist badges
Genome-wide visualization — pycirclize Circos plot (SVs, CNV gains/losses, STR expansions, SMN locus)
3-tier clinical HTML report — ACMG SF v3.2 Tier 1, OMIM morbid Tier 2, Tier 3 with full XLS export; Bootstrap 5, embedded SVG, optional truvari GIAB benchmark
Reproducible — every tool pinned in a Docker container; no conda required

Pipeline Overview

┌───────────────────────────────────────────────────────────────────────────────────┐
│ M1 · Pre-processing                                                               │
│ BWA-MEM2 → sort → MarkDup → mosdepth QC                                           │
└─────────────────────────────────────────┬─────────────────────────────────────────┘
                                          │ BAM
         ┌─────────────────────┬──────────┴─────────┬─────────────────────┐
         ▼                     ▼                    ▼                     ▼
┌──────────────────┐ ┌───────────────────┐ ┌──────────────────┐ ┌───────────────────┐
│ M2 · SV Calling  │ │ M3 · CNV + traits │ │ M4 · SMN Calling │ │ M8 · Alpha-globin │
│ Manta  DELLY     │ │ CNVpytor          │ │ SMNCopyNumber    │ │ HBA1/HBA2 depth   │
│ GRIDSS  Scramble │ │ GATK gCNV → BED   │ │ Caller v1.1      │ │ + junction search │
│ MELT → JASMINE   │ │ RHD/AMY1/GSTM1    │ │                  │ │ + targeted site   │
│ (5-caller merge) │ │ GSTT1/LPA KIV-2   │ │                  │ │ pileup            │
└────────┬─────────┘ └─────────┬─────────┘ └────────┬─────────┘ └─────────┬─────────┘
         │                     │                    │                     │
         ▼                     │                    │                     │
 ┌───────────────┐
 │ M5 · Annotate │             │                    │                     │
 │ AnnotSV 3.4   │             │                    │                     │
 │ gnomAD-SV AF  │             │                    │                     │
 │ filter <1%    │             │                    │                     │
 └───────┬───────┘             │                    │                     │
         └─────────────────────┴──────────┬─────────┴─────────────────────┘
                                          ▼
┌───────────────────────────────────────────────────────────────────────────────────┐
│ M6/M7 · Visualization & Reporting                                                 │
│ pycirclize Circos + Jinja2 HTML report + (optional) Truvari benchmark             │
└───────────────────────────────────────────────────────────────────────────────────┘

Requirements

Dependency	Version	Notes
Nextflow	≥ 25.10.4	`curl -s get.nextflow.io \| bash`
Docker	≥ 24	All tools run in containers
Python	≥ 3.11	For local test/report scripts
Java	≥ 17	Required by Nextflow

System: 16+ CPU cores, 64+ GB RAM, 500 GB disk recommended for 30× WGS.

Quick Start

# 1. Clone
git clone https://github.com/alvin8-git/SVcaller.git
cd SVcaller

# 2. Install Nextflow (if not already installed)
curl -s https://get.nextflow.io | bash
sudo mv nextflow /usr/local/bin/

# 3. Download reference data (~60 min first time)
bash validation/download_refs.sh

# 4. Run on a sample
nextflow run main.nf \
    -profile docker \
    --input /path/to/samplesheet.csv \
    --ref_fasta /path/to/ref/GRCh38/GRCh38.fasta \
    --outdir results

Input Samplesheet

A CSV file with one row per sample. Provide either fastq_1/fastq_2 or bam — not both.

sample,fastq_1,fastq_2,bam
HG002,/data/HG002_R1.fastq.gz,/data/HG002_R2.fastq.gz,
HG003,,,/data/HG003.GRCh38.bam

Column	Required	Description
`sample`	Yes	Unique sample ID (no spaces)
`fastq_1`	FASTQ mode	Absolute path to R1 FASTQ.gz
`fastq_2`	FASTQ mode	Absolute path to R2 FASTQ.gz
`bam`	BAM mode	Absolute path to sorted BAM

Parameters

Parameter	Default	Description
`--input`	required	Path to samplesheet CSV
`--ref_fasta`	required	GRCh38 reference FASTA
`--outdir`	`results`	Output directory
`--min_depth`	`25`	Minimum mean coverage (fails pipeline if below)
`--pon`	null	GATK gCNV Panel of Normals HDF5 (see PoN Build)
`--intervals`	null	Preprocessed intervals BED for GATK gCNV
`--annotsv_db`	null	AnnotSV annotation directory
`--eh_catalog`	`assets/eh_catalog.json`	ExpansionHunter variant catalog
`--sv_pon`	null	GIAB 7-sample SV Panel of Normals BED for artifact flagging (see `pon/sv_pon/`)
`--giab_truth`	null	GIAB T2TQ100-V1.0 truth VCF.gz for truvari benchmarking
`--giab_truth_v5q`	null	GIAB v5.0q truth VCF.gz (second benchmark pass)
`--skip_gridss`	`false`	Skip GRIDSS (saves 4–6 h; leaves Manta+DELLY+Scramble+MELT)
`--skip_melt`	`false`	Skip MELT MEI calling (saves ~2 h when container unavailable)
`--auto_cleanup`	`false`	Delete the `-work-dir` automatically on successful completion (removes the `-resume` cache)
`--max_cpus`	`64`	Max CPUs any process may request (caps `task.attempt` scaling)
`--max_memory`	`120.GB`	Max memory any process may request
`--max_time`	`240.h`	Max wall-clock time any process may request

Running the Pipeline

FASTQ input

NXF_ANSI_LOG=false nohup nextflow run main.nf \
    -profile docker \
    --input samplesheet.csv \
    --ref_fasta /path/to/hg38.canonical.fa \
    --pon /path/to/SVcaller/pon/pon/giab_cnv_pon.hdf5 \
    --annotsv_db /path/to/annotsv/Annotations_Human \
    --sv_pon /path/to/SVcaller/pon/sv_pon/giab_sv_pon.bed \
    --outdir results_SAMPLEID \
    -work-dir work_SAMPLEID \
    > /tmp/SAMPLEID_run1.log 2>&1 &

Note: Use hg38.canonical.fa (chr1-22+X+Y+M only) for FASTQ inputs — this skips FILTER_CHROMS (~70 min/sample on a ~30× BAM). Use NXF_ANSI_LOG=false for all background runs; without it, Nextflow's ANSI renderer deadlocks the JVM when there's no TTY.

BAM input (skip alignment)

NXF_ANSI_LOG=false nohup nextflow run main.nf \
    -profile docker \
    --input bam_samplesheet.csv \
    --ref_fasta /path/to/hg38.fa \
    --pon /path/to/SVcaller/pon/pon/giab_cnv_pon.hdf5 \
    --sv_pon /path/to/SVcaller/pon/sv_pon/giab_sv_pon.bed \
    --outdir results_SAMPLEID \
    -work-dir work_SAMPLEID \
    > /tmp/SAMPLEID_run1.log 2>&1 &

Note: BAM inputs always run FILTER_CHROMS to strip non-canonical @SQ headers. See How to run BAM inputs for details.

With GIAB benchmarking

NXF_ANSI_LOG=false nohup nextflow run main.nf \
    -profile docker \
    --input samplesheet.csv \
    --ref_fasta /path/to/hg38.canonical.fa \
    --giab_truth /path/to/GIAB/HG002_T2TQ100-V1.0_stvar.vcf.gz \
    --giab_truth_v5q /path/to/GIAB/HG002_v5.0q_stvar.vcf.gz \
    --outdir results \
    -work-dir work \
    > HG002_run1.log 2>&1 &

All samples share a single unified --outdir results (Nextflow writes each sample to its own results/<sample_id>/ subdir) and a single -work-dir work scratch directory. See Output for the consolidated layout.

Resume a failed run

NXF_ANSI_LOG=false nohup nextflow run main.nf \
    -profile docker \
    --input samplesheet.csv \
    --ref_fasta /path/to/hg38.canonical.fa \
    -work-dir work_SAMPLEID \
    -resume \
    > /tmp/SAMPLEID_run2.log 2>&1 &

Storage, Cleanup & Environment

A 30× WGS run can leave several hundred GB of intermediates in its work directory. Three hardening rules keep disk usage bounded:

One -work-dir per sample/batch — work_<sampleId> for single samples, work_<batchName> for batches. Never share a work/ across runs: it causes Nextflow session-lock conflicts and blocks targeted cleanup.
Clean up after results are published — bash bin/nf-cleanup.sh <sampleId> verifies outputs exist under --outdir, removes that sample's work dir, and prunes orphaned .nextflow/cache sessions. After several runs have shared one work dir, bash bin/nf-cleanup.sh --reclaim reclaims every superseded run's intermediates in one command (dry-run by default; add --force to delete) while keeping the latest successful run for -resume; it refuses to run while a pipeline is active. --auto_cleanup true deletes the work dir automatically on success (drops the -resume cache; one-shot runs only).
Keep the storeDir caches — ${outdir}/cache/ and ${outdir}/.cache/ persist GRIDSS reference setup, GATK interval binning, and chrom-filtered BAMs across runs against the same reference. They survive nextflow clean; don't delete them between samples.
Mosdepth writes regions only — --no-per-base is set, so the ~4.5 GB/sample per-base BED is never generated (the report consumes the 50 kb regions windows). Saves disk and mosdepth runtime.

Environment variables for background and multi-user runs:

Variable	Purpose
`NXF_ANSI_LOG=false`	Required for nohup/background runs — without it Nextflow's ANSI renderer deadlocks the JVM when there is no TTY.
`TMPDIR`	Redirect temp files off a small root partition (pipeline sets `/path/to/tmp` in `nextflow.config`; override per site).
`NXF_HOME`	Per-user Nextflow home on shared machines — avoids plugin-cache contention.
`NXF_SINGULARITY_CACHEDIR`	Per-user image cache for Singularity/Apptainer — stops concurrent runs colliding while pulling the same image.

Resource bounds & optimization

Per-process CPU/memory/time are capped by --max_cpus (64), --max_memory (120.GB), --max_time (240.h). Lower them on small machines (e.g. --max_cpus 8 --max_memory 32.GB) so no process over-subscribes. Tiers and per-caller overrides live in conf/base.config.

Scatter-gather is applied only where it is safe for SV/CNV calling:

Stage	Parallelization
DELLY	internal per-chromosome scatter-gather (built in)
GATK CollectReadCounts / CNVpytor	per-bin / per-chromosome — shardable
ExpansionHunter	per-locus
Manta / GRIDSS / SvABA	not chunked — breakend assembly needs whole-genome context; per-chromosome splitting drops inter-chromosomal events. Bound cost with `--tiered_gridss` or `--skip_gridss` instead.

scratch (node-local staging) is off by default. On a single-node local executor it doubles I/O on the ~130 GB filtered BAM for no benefit; enable it only on a shared-filesystem cluster via a site config (-c site.config setting process.scratch = true).

For per-process resource tiers, the local (conda) vs docker profiles, and the full storage/cache reference, see Parameter reference.

Panel of Normals

GATK gCNV requires a Panel of Normals (PoN) built from ≥10 normal samples. Build once from GIAB HG001–HG007:

# 1. Edit validation/giab_samplesheet.csv with your BAM paths
# 2. Build PoN
NXF_ANSI_LOG=false nohup nextflow run workflows/pon_build.nf \
    -profile docker \
    --input validation/giab_samplesheet.csv \
    --ref_fasta /path/to/ref/GRCh38/GRCh38.fasta \
    --outdir /path/to/SVcaller/pon \
    -work-dir /path/to/SVcaller/work_pon \
    > /path/to/tmp/pon_run.log 2>&1 &

# PoN output: /path/to/SVcaller/pon/pon/giab_cnv_pon.hdf5

Pass it to the main pipeline with --pon /path/to/SVcaller/pon/pon/giab_cnv_pon.hdf5.

Output

All runs share one consolidated results/ tree, one subdirectory per sample. Current occupants (see ORGANIZATION.md for the reorg history):

results/
├── HG002/                    # GIAB reference (SV/CNV/STR/SMN + blood-group traits)
├── HG001/                    # GIAB NA12878 (30X GRCh38 + blood-group traits)
├── COLO829/                  # tumor SV/CNV (depth traits omitted — no BAM)
├── SMN/{SMAD,SMAM,SMAPB}/    # SMA trio
├── _experiments/{HG002_wedge,svpon}/   # retained experimental runs
├── pon/                      # shared Panel of Normals (single copy)
└── cache/                    # shared reference-derived pipeline cache (gitignored)

Each sample subdirectory contains:

results/
└── <sample_id>/
    ├── <sample_id>.report.html          # Self-contained HTML report
    ├── <sample_id>.variants.xlsx        # Excel workbook (SVs / CNVs / STRs / SMN sheets)
    ├── <sample_id>.sv_merged.vcf.gz     # Ensemble SV calls (Manta+DELLY+GRIDSS+Scramble+MELT)
    ├── <sample_id>.sv_merged.vcf.gz.tbi
    ├── <sample_id>.str.vcf.gz           # STR calls (ExpansionHunter)
    ├── <sample_id>.cnv_consensus.bed    # Consensus CNV calls
    ├── <sample_id>.smn.tsv              # SMN1/SMN2 copy numbers
    ├── <sample_id>.filtered.tsv         # AnnotSV-annotated, gnomAD-SV-filtered SVs
    ├── <sample_id>.circos.svg           # Genome-wide Circos plot (embedded inline in HTML)
    ├── <sample_id>.circos.png
    ├── bloodgroup/
    │   └── <sample_id>.rh_status.tsv    # Rh factor: RHD copy number → Rh+/- (CNV trait)
    ├── cnv_traits/
    │   ├── <sample_id>.amy1.tsv         # AMY1 (salivary amylase) copy number
    │   ├── <sample_id>.gst_null.tsv     # GSTM1 / GSTT1 null (homozygous deletion) status
    │   └── <sample_id>.lpa_kiv2.tsv     # LPA KIV-2 tandem-repeat copy number (Lp(a))
    ├── alpha_globin/                    # M8 · alpha-globin (HBA1/HBA2)
    │   ├── <sample_id>.alpha_globin.tsv # contract: dosage, deletion group, site genotypes
    │   ├── <sample_id>.alpha_depth.tsv  # per-segment normalized depth
    │   ├── <sample_id>.alpha_junction.tsv # alpha-cluster breakpoint search
    │   └── <sample_id>.alpha_sites.tsv  # pinned pathogenic-site pileup
    ├── qc/                              # published alignment QC (survives work-dir cleanup)
    │   ├── <sample_id>.flagstat.txt     # mapping rate (true, measured pre-FILTER_CHROMS)
    │   ├── <sample_id>.mosdepth.summary.txt
    │   └── <sample_id>.insert_size_metrics.txt
    └── <sample_id>.truvari/             # GIAB benchmark (if --giab_truth set)
        └── summary.json

Copy-number / blood-group trait contracts

Small, stable per-sample TSVs emitted by the CNV_TRAITS subworkflow for downstream interpretation (e.g. OmniGen). Each is a targeted, normalized read-depth estimate from the aligned BAM (mosdepth over assets/cnv_trait_regions.bed, --mapq 0, normalized to copy-number-stable CTRL_* control regions); the CNV consensus BED corroborates called deletions but does not drive the call.

File	Columns
`bloodgroup/<S>.rh_status.tsv`	`sample`, `RHD_copies`, `Rh_status` (pos/neg), `confidence`
`cnv_traits/<S>.amy1.tsv`	`sample`, `AMY1_copies`, `method`
`cnv_traits/<S>.gst_null.tsv`	`sample`, `GSTM1` (null/present), `GSTT1` (null/present)
`cnv_traits/<S>.lpa_kiv2.tsv`	`sample`, `KIV2_copies`, `method`

See docs/omnigen-additions-plan.md for design, coordinates, thresholds, and calibration notes.

The HTML report includes:

Alignment QC (coverage, duplication rate, insert size)
SV summary table by type and caller
Embedded Circos plot (SVs, CNV gains/losses, STR expansions, SMN locus, and a labelled CNV-trait ring — RHD/AMY1/GSTM1/GSTT1/LPA KIV-2 — when trait TSVs are present)
SMN1/SMN2 copy number with SMA carrier/affected classification
Tier 1 — ACMG SF v3.2 actionable SVs (≥2 callers, no PON hit)
Tier 2 — OMIM morbid gene candidates, top 10 shown; full list in XLS download
Tier 3 — all remaining SVs, top 10 shown; full list in XLS download
STR expansion loci with INREPEAT / NORMAL / INTERMEDIATE status
Blood Group & Copy-Number Traits card — Rh(D) factor (RHD copy number), AMY1 copies, GSTM1/GSTT1 null status, LPA KIV-2 copies
Alpha-globin (HBA1/HBA2) card — alpha-gene dosage (0–4), genotype, deletion group (e.g. --SEA|--MED when depth cannot separate the alleles), and panel-named pathogenic-site findings. Measurement only, no clinical call
GIAB benchmark precision/recall/F1 with per-size-bin breakdown (optional)

GIAB Benchmarking

Validate SV calls against GIAB truth sets:

bash validation/giab_benchmark.sh \
    HG002 \
    results/HG002/HG002.sv_merged.vcf.gz

Current benchmark on HG002 (run16, 2026-06-08; GRIDSS BND→SV fix). These numbers predate two caller fixes and are effectively a 4-caller result. SvABA was never merged (vcfs[5] is unwired in modules/jasmine/merge.nf), and MELT was silently failing on an invalid -reads flag from 2026-06-05 until the -sr fix on 2026-07-23, so only Manta + DELLY + GRIDSS + Scramble contributed. Keep this in mind before quoting it as a 5-caller F1; a re-benchmark with MELT contributing is pending:

Benchmark	Precision	Recall	F1	TP-base	FP
GIAB T2TQ100-V1.0	0.733	0.255	0.378	7571	2604
GIAB v5.0q	0.738	0.259	0.383	7286	2449

Main gap is recall (~25%) — the callset covers ~9,700 of ~30,000 truth SVs. Precision is solid. Next target: low-QUAL Manta/Delly rescue or soft GRIDSS QUAL floor.

Demo report

A complete, self-contained per-sample report is committed at docs/demo/HG002_report.html — open it in any browser (Bootstrap CSS and Circos SVG are inlined; no network needed).

It is rendered by bin/html_report.py from HG002's real existing outputs (SV tiers from HG002.filtered.tsv, HG002.cnv_consensus.bed, HG002.smn.tsv, embedded Circos plot). The Blood Group & Copy-Number Traits card (Rh(D) factor, AMY1 copies, GSTM1/GSTT1 null, LPA KIV-2) now shows real, validated values computed from targeted normalized read depth on the 30X HG002 BAM (2026-07-12): Rh(D) positive (RHD CN 2), GSTM1/GSTT1 present, AMY1 2, LPA KIV-2 5. The Rh(D)-positive call agrees with the CNV consensus (no RHD deletion) and the GSTM1 het-deletion call agrees with GIAB v5.0q truth — see the validation table in docs/omnigen-additions-plan.md. AMY1 and LPA KIV-2 absolute copies remain proportional-only pending calibration.

As of 2026-07-13 the demo again carries every section, including the two whose inputs were lost when the work_HG002 scratch dir was reclaimed (both stages were re-run, not reconstructed):

Short Tandem Repeat (STR) Findings — ExpansionHunter 5.0.0 over the 32-locus assets/eh_catalog.json.
Validation Benchmark — Truvari 4.3.1 against both GIAB HG002 SV truth sets. On this Manta+DELLY call set (--skip_gridss): T2TQ100-V1.0 precision 0.733 / recall 0.255 / F1 0.378 (TP 7,135 · FP 2,605 · FN 22,120); v5.0q precision 0.738 / recall 0.259 / F1 0.383 (TP 6,884 · FP 2,450 · FN 20,837). Recall is bounded by short reads against an INS-heavy assembly truth set, with per-size-bin metrics in the report.

Two more full reports are committed alongside it:

docs/demo/HG001_report.html — GIAB NA12878, regenerated from real 30X outputs with the trait card computed from the HG001 BAM (Rh(D) positive, GSTM1/GSTT1 present, AMY1 2, KIV-2 6). Note: NA12878 is often cited as a "GSTM1-null" anchor, but this BAM reads ~1 copy at GSTM1 (het deletion), not a homozygous null — see the validation table in docs/omnigen-additions-plan.md.
docs/demo/COLO829_report.html — COLO829 tumor, SV/CNV focus. The Blood Group & Copy-Number Traits card carries an explicit note that depth traits were not computed (no aligned BAM was available).

Documentation

Document	What it covers
Getting started tutorial	First run from scratch to open report
How to run a clinical sample	End-to-end guide for a new patient sample (samplesheet → results → cleanup)
How to run BAM inputs	Pre-aligned BAM guidance including FILTER_CHROMS verification and parallel runs
How to run the GIAB validation	Benchmarking against GIAB HG002 truth set
How to build a Panel of Normals	GATK gCNV PON construction
How to interpret the HTML report	Clinical interpretation of all report sections and Circos rings
Parameter reference	All CLI flags, samplesheet format, output files, resource labels
Architecture reference	Module-by-module technical description with I/O and design notes
Design decisions	Why the pipeline is built the way it is (channel patterns, sentinel files, PON choices, FILTER_CHROMS @SQ fix, no-empty-placeholder rule)
Changes	Notable behavioural changes and the incidents that motivated them

Failure semantics — no empty placeholders

A failed caller never publishes an output file. If a stage cannot produce a real result, it exits non-zero and the run stops (errorStrategy = 'finish', see conf/base.config). SVcaller will not write a zero-byte or fabricated-empty artifact to results/.

This matters because a downstream consumer cannot distinguish the two on disk. An earlier version of this pipeline used || touch <output> fallbacks, so a crashed SMN caller published an empty smn.tsv, exited 0, and a consumer report rendered it as "0 Carrier findings — Clear" for a real person. See docs/CHANGES.md for the full incident write-up.

The distinction the pipeline maintains:

A legitimate empty result is still published. A VCF with a proper header and zero variant rows means "we looked and found nothing" — that is a real, valid answer.
An intentionally skipped stage is explicit. Use --skip_melt, --skip_scramble, --skip_svaba or --skip_gridss. These route to a dedicated *_STUB process that emits a well-formed header-only VCF. This is the only sanctioned way to run without a caller.
A crashed or missing caller is a hard failure. It is never silently downgraded to an empty result.

Practical consequence: if MELT is not installed, MELT_CALL now fails with an error telling you to install it or pass --skip_melt — it no longer emits an empty VCF that reads as "no mobile element insertions found".

Testing

# Install test dependencies
pip install pytest pycirclize matplotlib jinja2 openpyxl pandas

# Run all unit tests
pytest tests/ -v

Tests cover: samplesheet validation, CNV consensus merging, SMN classification, Circos plot parsing, HTML report rendering, and CNV/blood-group trait interpreters (Rh/RHD, AMY1, GSTM1/GSTT1 null, LPA KIV-2 — arithmetic on synthetic depth fixtures, consensus cross-check against the real HG002 consensus BED, and contract-schema round-trips).

tests/test_no_empty_placeholders.py is a regression guard for the empty-placeholder incident described above: it statically asserts that no module reintroduces an || touch <output> fallback, an echo '{}' > placeholder, or a || true that swallows a caller's exit code.

Docker Containers

All tools run in pinned Docker containers. No local tool installation required beyond Docker itself.

Module	Container
BWA-MEM2	`quay.io/biocontainers/bwa-mem2:2.2.1--he70b90d_8`
samtools	`quay.io/biocontainers/samtools:1.23.1--ha83d96e_0`
Picard MarkDup	`quay.io/biocontainers/picard:3.2.0--hdfd78af_0`
mosdepth	`quay.io/biocontainers/mosdepth:0.3.14--h05c3d44_0`
FastQC	`quay.io/biocontainers/fastqc:0.12.1--hdfd78af_0`
Manta	`quay.io/biocontainers/manta:1.6.0--py27h9948957_6`
DELLY	`quay.io/biocontainers/delly:1.2.6--hdcf5f25_4`
GRIDSS	`gridss/gridss:2.13.2`
Scramble	`quay.io/biocontainers/scramble:1.0.2--h031d066_1`
MELT	`svcaller/melt:2.2.2` (local build — requires registration)
SvABA	`quay.io/biocontainers/svaba:1.2.0--h69ac913_1`
ExpansionHunter	`quay.io/biocontainers/expansionhunter:5.0.0--hc26b3af_5`
STRling	`quay.io/biocontainers/strling:0.5.2--hbbffb53_1`
JASMINE	`quay.io/biocontainers/jasminesv:1.1.5--hdfd78af_0`
CNVpytor	`quay.io/biocontainers/cnvpytor:1.3.1--pyhdfd78af_1`
GATK	`broadinstitute/gatk:4.5.0.0`
SMNCopyNumberCaller	`svcaller/smncopynum:1.1`
AnnotSV	`quay.io/biocontainers/annotsv:3.4.6--py313hdfd78af_0`
truvari	`quay.io/biocontainers/truvari:4.3.1--pyhdfd78af_0`
Python utils / report	`svcaller/utils:1.3` (local build)

Genome Build

GRCh38 / hg38 only. chr-prefixed chromosome names required.

License

MIT — see LICENSE.

Citation

If you use SVcaller in your research, please cite the individual tools (Manta, DELLY, GRIDSS, CNVpytor, GATK, SMNCopyNumberCaller, AnnotSV, ExpansionHunter) as appropriate for your analysis.

Name		Name	Last commit message	Last commit date
assets		assets
bin		bin
conf		conf
docs		docs
modules		modules
nxf-17344157757238233620		nxf-17344157757238233620
nxf-7620641549738606408		nxf-7620641549738606408
subworkflows		subworkflows
tests		tests
validation		validation
workflows		workflows
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
Dockerfile.melt		Dockerfile.melt
Dockerfile.smncopynum		Dockerfile.smncopynum
Dockerfile.utils		Dockerfile.utils
Documentation.md		Documentation.md
ORGANIZATION.md		ORGANIZATION.md
README.md		README.md
TODO.md		TODO.md
VERSION.md		VERSION.md
environment.yml		environment.yml
libjansi-64.so		libjansi-64.so
main.nf		main.nf
nextflow.config		nextflow.config

Navigation Menu

Repository files navigation

SVcaller

Features

Pipeline Overview

Requirements

Quick Start

Input Samplesheet

Parameters

Running the Pipeline

FASTQ input

BAM input (skip alignment)

With GIAB benchmarking

Resume a failed run

Storage, Cleanup & Environment

Resource bounds & optimization

Panel of Normals

Output

Copy-number / blood-group trait contracts

GIAB Benchmarking

Demo report

Documentation

Failure semantics — no empty placeholders

Testing

Docker Containers

Genome Build

License

Citation

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

SVcaller

Features

Pipeline Overview

Requirements

Quick Start

Input Samplesheet

Parameters

Running the Pipeline

FASTQ input

BAM input (skip alignment)

With GIAB benchmarking

Resume a failed run

Storage, Cleanup & Environment

Resource bounds & optimization

Panel of Normals

Output

Copy-number / blood-group trait contracts

GIAB Benchmarking

Demo report

Documentation

Failure semantics — no empty placeholders

Testing

Docker Containers

Genome Build

License

Citation

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages