Nextflow Workflow

This guide covers running py-gbcms as a Nextflow workflow for processing multiple samples in parallel, particularly on HPC clusters.

Overview

The Nextflow workflow provides:

• Automatic parallelization across samples

• SLURM/HPC integration with resource management

• Containerization with Docker/Singularity

• Resume capability for failed runs

• Reproducible pipelines

Prerequisites

1. Nextflow >= 21.10.3

2. One of:

   • Docker (for local)

   • Singularity (for HPC)

Install Nextflow:

curl -s https://get.nextflow.io | bash
mv nextflow ~/bin/  # or any directory in your PATH

Quick Start

1. Prepare Samplesheet

Create a CSV file with your samples:

sample,bam,bai
sample1,/path/to/sample1.bam,/path/to/sample1.bam.bai
sample2,/path/to/sample2.bam,
sample3,/path/to/sample3.bam,/path/to/sample3.bam.bai

Or with per-sample suffix (for multiple BAM types):

sample,bam,bai,suffix
sample1,/path/to/sample1.duplex.bam,,-duplex
sample1,/path/to/sample1.simplex.bam,,-simplex
sample1,/path/to/sample1.unfiltered.bam,,-unfiltered
sample2,/path/to/sample2.bam,,

Notes:

• bai column is optional - will auto-discover <bam>.bai if not provided

• suffix column is optional - per-row suffix overrides global --suffix parameter

• BAI files must exist or workflow will fail early with clear error

2. Run the Workflow

Local with Docker:

nextflow run nextflow/main.nf \
    --input samplesheet.csv \
    --variants variants.vcf \
    --fasta reference.fa \
    --outdir results \
    -profile docker

SLURM cluster with Singularity:

nextflow run nextflow/main.nf \
    --input samplesheet.csv \
    --variants variants.vcf \
    --fasta reference.fa \
    --outdir results \
    -profile slurm

Parameters

Required

Parameter	Description
--input	Path to samplesheet CSV
--variants	Path to VCF/MAF variants file
--fasta	Reference FASTA (with .fai index)

Output Options

Parameter	Default	Description
--outdir	results	Output directory
--format	vcf	Output format (vcf or maf)
--suffix	''	Suffix for output filenames

Filtering Options

Parameter	Default	Description
--min_mapq	20	Minimum mapping quality
--min_baseq	0	Minimum base quality
--filter_duplicates	true	Filter duplicate reads
--filter_secondary	false	Filter secondary alignments
--filter_supplementary	false	Filter supplementary alignments
--filter_qc_failed	false	Filter QC failed reads
--filter_improper_pair	false	Filter improperly paired reads
--filter_indel	false	Filter reads with indels

Resource Limits

Parameter	Default	Description
--max_cpus	16	Maximum CPUs per job
--max_memory	128.GB	Maximum memory per job
--max_time	240.h	Maximum runtime per job

Execution Profiles

Docker (Local)

-profile docker

• Uses Docker containers

• Best for local development

• Requires Docker installed

Singularity (HPC)

-profile singularity

• Uses Singularity images

• Best for HPC without SLURM

• Requires Singularity installed

SLURM (HPC Cluster)

-profile slurm

• Submits jobs to SLURM

• Uses Singularity containers

• Queue: cmobic_cpu (customizable)

Customizing for Your Cluster

Edit nextflow/nextflow.config to customize the SLURM profile:

slurm {
    process.executor       = 'slurm'
    process.queue          = 'your_queue_name'  // Change this
    process.clusterOptions = '--account=your_account'  // Add if needed
    singularity.enabled    = true
    singularity.autoMounts = true
}

Common customizations:

process {
    withName: GBCMS_RUN {
        cpus   = 8          // CPUs per sample
        memory = 16.GB      // Memory per sample
        time   = 6.h        // Time limit per sample
    }
}

Output Structure

Results are organized in ${outdir}/:

results/
├── gbcms/
│   ├── sample1.vcf        # Or .maf
│   ├── sample2.vcf
│   └── sample3.vcf
└── pipeline_info/
    ├── execution_report.html
    ├── execution_timeline.html
    └── execution_trace.txt

Advanced Usage

Resume Failed Runs

Nextflow caches completed tasks. Resume from where it failed:

nextflow run nextflow/main.nf \
    --input samplesheet.csv \
    --variants variants.vcf \
    --fasta reference.fa \
    -profile slurm \
    -resume

Custom Suffix

Add suffix to output filenames:

--suffix .genotyped
# Output: sample1.genotyped.vcf

MAF Output

Generate MAF instead of VCF:

--format maf
# Output: sample1.maf

Strict Filtering

Enable all filters for high-quality genotyping:

nextflow run nextflow/main.nf \
    --input samplesheet.csv \
    --variants variants.vcf \
    --fasta reference.fa \
    --filter_duplicates true \
    --filter_secondary true \
    --filter_supplementary true \
    --filter_qc_failed true \
    -profile slurm

Monitoring

View Running Jobs

# SLURM
squeue -u $USER

# Nextflow
nextflow log

Check Progress

Nextflow prints real-time progress:

[c3/a1b2c3] GBCMS_RUN (sample1) [100%] 10 of 10 ✔

Execution Report

After completion, view the HTML report:

open results/pipeline_info/execution_report.html

Troubleshooting

Job Failed with Error

Check the work directory in error message:

cat work/c3/a1b2c3/.command.log

Out of Memory

Increase memory in config:

process {
    withName: GBCMS_RUN {
        memory = 32.GB
    }
}

Wrong Queue

Update queue name in nextflow/nextflow.config:

process.queue = 'your_queue_name'

Missing Container

Pull the container manually:

# Singularity
singularity pull docker://ghcr.io/msk-access/py-gbcms:2.0.0

# Docker
docker pull ghcr.io/msk-access/py-gbcms:2.0.0

Comparison with CLI

Feature	CLI	Nextflow
Multiple samples	Sequential	Parallel
Resource management	Manual	Automatic
Retry failed jobs	Manual	Automatic
HPC integration	Manual scripts	Built-in
Resume capability	No	Yes

When to use CLI instead: See Usage Patterns

Next Steps

• See Usage Patterns for comparison with CLI usage

• See nextflow/README.md for additional workflow documentation