nl-rna-varcall

A Nextflow pipeline for variant calling from RNA-seq aligned data using DeepVariant with customized models for transcriptomic data.

Overview

This pipeline processes RNA-seq alignment files (BAM/CRAM) to identify genetic variants using a three-step approach:

1. Coverage Analysis (Mosdepth): Calculate per-base coverage depth
2. Region Filtering (BedTools): Intersect high-coverage regions with coding sequences
3. Variant Calling (DeepVariant): Call variants using a custom RNA-seq model

Workflow

flowchart TD
    A[RNA-seq Alignment<br/>BAM/CRAM] --> B[MOSDEPTH<br/>Coverage Analysis]
    C[Reference Genome<br/>FASTA + FAI] --> B
    B --> D[BEDTOOLS_MERGE_INTERSECT<br/>Filter High-Coverage CDS]
    E[GENCODE CDS BED] --> D
    D --> F[DEEPVARIANT_RUNDEEPVARIANT<br/>Variant Calling]
    A --> F
    C --> F
    G[Custom DeepVariant Model] --> F
    F --> H[VCF + gVCF Output]

Loading

Requirements

• Nextflow >= 22.04.0
• Docker (enabled)
• Conda (enabled)

System Requirements

• Memory: 192 GB for DeepVariant process
• CPUs: 48 cores for DeepVariant process
• Storage: Sufficient space for intermediate files and outputs

Configuration

Required Parameters

Parameter	Description	Example
sample_name	Sample identifier	"UDN486800-2931649-MGML0089-FBR1"
alignment	Path to BAM/CRAM file	"/path/to/sample.cram"
alignment_index	Path to BAM/CRAM index	"/path/to/sample.cram.crai"
fasta	Reference genome FASTA	"/path/to/GRCh38.fa"
fai	Reference genome index	"/path/to/GRCh38.fa.fai"
gencode_bed	GENCODE CDS BED file	"/path/to/gencode.cds.bed"
min_coverage	Minimum coverage threshold	3

DeepVariant Model Files

Parameter	Description	Example
model_data	Model data file	"/path/to/model.ckpt.data-00000-of-00001"
model_index	Model index file	"/path/to/model.ckpt.index"
model_meta	Model metadata file	"/path/to/model.ckpt.meta"
model_info	Model info file	"/path/to/model.ckpt.example_info.json"

Docker Images

Parameter	Description
mosdepth_docker	Mosdepth container image URI
bedtools_docker	BedTools container image URI
deepvariant_docker	DeepVariant container image URI

Outputs

The pipeline generates the following outputs in the specified output directory:

BED Files (/BED/)

• Coverage BED: Per-base coverage from Mosdepth
• Filtered CDS BED: High-coverage coding sequences

Variant Calls (/VARCALL/)

• VCF file (*.vcf.gz): Compressed variant calls
• VCF index (*.vcf.gz.tbi): Index for VCF file
• gVCF file (*.g.vcf.gz): Genomic VCF with all sites
• gVCF index (*.g.vcf.gz.tbi): Index for gVCF file
• HTML Report (*.visual_report.html): DeepVariant visual report

AWS HealthOmics Deployment

Create Pipeline Package

git clone https://github.com/uclanelsonlab/nl-rna-varcall.git 
cd nl-rna-varcall/

# Create deployment package
zip -r nl-rna-varcall.zip *

Test Data

You can validate the pipeline using these reference samples:

# RNA-seq sample
s3://gatk-test-data/rna_bam/NA12878_b37/NA12878.bam
s3://gatk-test-data/rna_bam/NA12878_b37/NA12878.bam.bai

# Reference files (hg19/b37)
s3://broad-references/hg19/v0/human_g1k_v37_decoy.fasta
s3://broad-references/hg19/v0/human_g1k_v37_decoy.fasta.fai
s3://broad-references/hg19/v0/human_g1k_v37_decoy.dict

Pipeline Architecture

Modules

• modules/mosdepth/: Coverage depth calculation
• modules/bedtools/: BED file operations and filtering
• modules/deepvariant/: Variant calling with custom models

Key Features

• Custom DeepVariant Models: Optimized for RNA-seq data
• Coverage-based Filtering: Focus on high-confidence regions
• CDS-specific Analysis: Target coding sequences for variant calling
• Scalable Processing: Configurable CPU/memory allocation
• Docker Integration: Consistent execution environments

Troubleshooting

Common Issues

1. Model File Staging: Ensure all DeepVariant model files are accessible
2. Memory Requirements: DeepVariant requires substantial memory (192GB)
3. File Permissions: Check read permissions for all input files
4. Container Access: Verify Docker images are accessible

Debugging

Check Nextflow logs for detailed error information:

nextflow log

Examine work directories for process-specific errors:

ls -la work/

License

This project is licensed under the terms specified in the LICENSE file.