🧬 Nanopore SV Duplication Detection Pipeline

A complete pipeline for structural variant discovery, focusing on duplications (SVTYPE=DUP), using Oxford Nanopore data. The workflow includes read preprocessing, mapping, SV calling (8 tools), and performance reporting. Designed for large-scale SVs, with optimized parameters for duplications.

---

🚀 Overview

This pipeline:

1. Downloads Nanopore reads using SRA tools.
2. Trims and filters reads for quality.
3. Aligns reads to a reference genome with minimap2.
4. Calls structural variants using eight SV callers:
   • Sniffles1
   • Sniffles2
   • cuteSV
   • DeBreak
   • Dysgu
   • SVIM
   • pbsv
   • SVDSS
5. Measures performance (time, memory, CPU) for each tool.
6. Produces VCF files and indexed BAMs.
7. Supports downstream benchmarking with Truvari.

---

🛠 Requirements

• Unix/Linux system
• conda (with conda-forge + bioconda)
• Tools (installed in dedicated conda environments):

Task	Tool(s) Used
Download	prefetch, fasterq-dump (from sra-tools)
QC	NanoPlot, NanoFilt, Porechop
Mapping	minimap2, samtools
SV Calling	sniffles, cuteSV, debreak, svim, pbsv, dysgu, SVDSS
Evaluation	Truvari, bcftools

---

📁 Inputs

• sample.txt: One SRA accession per line.
• Reference genome: FASTA format (*.fasta or *.fna).
  • Make sure it's indexed if needed (samtools faidx, minimap2 -d).
• *.svsig.gz files are reused for pbsv to avoid repeating discover.

---

🧬 Pipeline Structure

Step 1: Data Download and Preprocessing

• Download reads with prefetch, convert with fasterq-dump.
• Combine and QC reads (NanoPlot, Porechop, NanoFilt).
• Trimming: adapters + low Q-score filtering.

Step 2: Mapping

• Align reads using minimap2 -x map-ont.
• Convert, sort, index BAM files.
• Generate per-base coverage.

Step 3: Structural Variant Calling

The pipeline uses 8 state-of-the-art SV callers, each with unique algorithms and strengths:

Tool	Description
Sniffles1	One of the first SV callers for long reads; identifies SVs using split-read and coverage signals.
Sniffles2	Improved version of Sniffles; supports better genotyping and is optimized for high-throughput data.
cuteSV	Fast and memory-efficient SV caller that clusters signals and refines SV boundaries.
DeBreak	Accurate SV caller using partial order alignment and duplication recovery for noisy long reads.
Dysgu	Versatile SV tool supporting many sequencing types; efficient with nanopore data.
SVIM	Detects a wide range of SVs using a signal-based approach from long-read alignments.
pbsv	PacBio’s official SV caller, works well with aligned reads including ONT; supports split-read analysis.
SVDSS	High-resolution caller using smoothing and suffix-filtering strategies to detect SVs with precision.

All tools support multi-threading and are configured for maximum SV size where applicable.

Step 4: Performance Logging

Each SV caller logs:

• Elapsed wall time
• CPU usage
• Peak memory
• User/system time

Results are saved in <SAMPLE>_performance_report.csv.

---

📤 Output Per Sample

File/Folder	Description
*_mapped.sort.bam + .bai	Aligned BAM and index
*_coverage.txt	Per-base coverage
*_caller.vcf	VCF output from each SV tool
<tool>_<SAMPLE>/	Tool-specific outputs (e.g., svim/, pbsv/)
*_performance_report.csv	Time, memory, CPU stats per caller
nanoplot_<SAMPLE>/	Quality metrics and plots

---

🔍 Post-processing (optional)

• Run truvari to compare output to golden standard.
• Filter only SVTYPE=DUP from VCFs using bcftools view -i 'INFO/SVTYPE=="DUP"'.

---

📚 References

• Sniffles
• cuteSV
• DeBreak
• SVIM
• pbsv
• Dysgu
• SVDSS
• Truvari
• NanoPlot
• Porechop
• NanoFilt

---

💡 Tips

• Run on an HPC cluster or use GNU parallel to scale samples.
• Ensure your reference genome and read quality are appropriate.
• Tune parameters (e.g. --max_size) depending on expected SV length.

---

🧪 Status

✅ Bash version: fully tested
⚠️ Python version: still under testing