nf-vcf2maf: Annotate VCF Files and Generate MAF Files

Purpose

The purpose of this Nextflow workflow is to annotate variants in VCF files using the Ensembl Variant Effect Predictor (VEP) and convert the annotated VCF file into the Mutation Annotation Format (MAF). Unlike VEP-anntotated VCF files, MAF files are generally more useful for downstream applications given their tabular nature. For example, you can easily load them into R (e.g. with the maftools package) and/or derive the input files for cBioPortal.

Important: Please read the limitations listed below.

This repository leverages a fork of vcf2maf and a custom container image.

Quickstart

1. Prepare a CSV samplesheet according to the format described below.

   Example: Stored locally as ./samplesheet.csv

   synapse_id  ,sample_parent_id ,merged_parent_id ,study_id ,variant_class ,variant_caller ,is_releasable
   syn87654301 ,syn87654311      ,syn87654321      ,study_x  ,germline      ,deepvariant    ,true
   syn87654302 ,syn87654311      ,syn87654321      ,study_x  ,germline      ,deepvariant    ,false
   syn87654303 ,syn87654311      ,syn87654321      ,study_x  ,germline      ,deepvariant    ,true
   syn87654304 ,syn87654312      ,syn87654321      ,study_x  ,germline      ,mutect2        ,false
   syn87654305 ,syn87654312      ,syn87654321      ,study_x  ,germline      ,mutect2        ,false
   syn87654306 ,syn87654312      ,syn87654321      ,study_x  ,germline      ,mutect2        ,false
   syn87654307 ,syn87654313      ,syn87654322      ,study_y  ,germline      ,deepvariant    ,true
   syn36245848 ,syn87654313      ,syn87654322      ,study_y  ,germline      ,deepvariant    ,true
   

2. Create a Nextflow secret called SYNAPSE_AUTH_TOKEN with a Synapse personal access token (docs).

3. Prepare your parameters file. For more details, check out the Parameters section. Only the input parameters is required.

   Example: Stored locally as ./params.yml

   input: "./samplesheet.csv"
   maf_center: "Sage Bionetworks"
   max_subpop_af: 0.0005

4. Launch workflow using the Nextflow CLI, the Tower CLI, or the Tower web UI.

   Example: Launched using the Nextflow CLI with Docker enabled

   nextflow run sage-bionetworks-workflows/nf-vcf2maf -params-file ./params.yml -profile docker

5. Explore the MAf files uploaded to Synapse (using the parent IDs listed in the samplesheet).

Authentication

This workflow takes care of transferring files to and from Synapse. Hence, it requires a secret with a personal access token for authentication. To configure Nextflow with such a token, follow these steps:

1. Generate a personal access token (PAT) on Synapse using this dashboard. Make sure to enable the view, download, and modify scopes since this workflow both downloads and uploads to Synapse.
2. Create a secret called SYNAPSE_AUTH_TOKEN containing a Synapse personal access token using the Nextflow CLI or Nextflow Tower.
3. (Tower only) When launching the workflow, include the SYNAPSE_AUTH_TOKEN as a pipeline secret from either your user or workspace secrets.

Parameters

Check out the Quickstart section for example parameter values. You are encouraged to read the limitations listed below because some parameters have not been tested with non-default values.

• input: (Required) A CSV samplesheet that lists the VCF files that should be processed. See below for the samplesheet format.
• max_subpop_af: Threshold used by vcf2maf for labeling variants with the common_variant filter. Specifically, the common_variant filter is applied to variants with an allele frequency of at least max_subpop_af in one or more gnomAD sub-populations (source). This filter is useful for removing false-positive somatic variants. The merged MAF files lack these common variants. Default: 0.0005.
• maf_center: Value used in the Center MAF column. Default: "Sage Bionetworks".
• reference_fasta: Reference genome FASTA file used in variant calling. Default: "s3://sage-igenomes/igenomes/Homo_sapiens/GATK/GRCh38/Sequence/WholeGenomeFasta/Homo_sapiens_assembly38.fasta".
• reference_fasta_fai: Reference genome FASTA index (FAI) file. This shouldn't be needed in most cases since the workflow will automatically pick up on the .fai file alongside the .fasta file. Default: "${reference_fasta}.fai".
• vep_tarball: A tarball (ideally compressed) of the VEP cache. Default: "s3://sage-igenomes/vep_cache/homo_sapiens_vep_107_GRCh38.tar.gz".
• ncbi_build: The NCBI genome build. Passed to --assembly in VEP (source). Default: "GRCh38".
• species: The species identifier. Passed to --species in VEP (source). Default: "homo_sapiens".

Inputs

Samplesheet

The input samplesheet should be in comma-separated values (CSV) format and contain the following columns. You should avoid using spaces or special characters in any of the columns. Otherwise, you might run into job caching issues.

1. synapse_id: Synapse ID of the VCF file
   • Make sure that the Synapse account associated with the personal access token has access to all listed VCF files
2. biospecimen_id: Biospecimen/sample identifier
   • This value will be used to populate the Tumor_Sample_Barcode MAF column
   • Important: This value needs to uniquely identify samples within each merged MAF file. See below for information on how MAF files are merged.
3. sample_parent_id: Synapse ID of the folder where the individual sample MAF file will be uploaded
   • Suggestion: The folder that contains the VCF file
4. merged_parent_id: The Synapse ID of the folder where the merged MAF file will be uploaded
   • Suggestion: The root folder containing the VCF files
   • Important: This value should be consistent across VCF files that are expected to be merged. Otherwise, you will end up with artificially split merged MAF files. See below for information on how MAF files are merged.
5. study_id: Study identifier
   • Suggestion: The Synapse ID of the project representing the study if you don’t have shorthand study IDs
6. variant_class: Whether the VCF file contains somatic or germline mutations
   • Valid values: somatic or germline
7. variant_caller: Name of the variant caller
8. is_releasable: Whether the VCF file should be included in the merged MAF file
   • Valid values: true or false

Outputs

MAF Files

• Individual sample MAF files
  • Unfiltered (i.e. includes all variants regardless of their FILTER status, including those that weren’t deemed high-confidence by the variant caller)
  • File naming: ${biospecimen_id}-${variant_class}-${variant_caller}.maf
• Merged study MAF files (one for every combination of study_id, variant_class and variant_caller)
  • Filtered (i.e. restricted to “releasable” samples and variants where FILTER == 'PASS', which excludes those with common_variant due to any(gnomAD_*_AF) >= 0.0005)
  • File naming: ${study_id}-${variant_class}-${variant_caller}.merged.maf

Known Limitations

• This workflow has only been tested with the following parameters:
  • vep_tarball: Ensembl VEP 107
  • species: homo_sapiens
  • ncbi_build: GRCh38
  • reference_fasta: GATK FASTA file

Benchmarks

Setup

The following benchmarks were performed with the following setup on an EC2 instance:

# Install tmux for long-running commands
sudo yum install -y tmux

# Install and setup Nextflow
sudo yum install -y java
(cd .local/bin && wget -qO- https://get.nextflow.io | bash)
echo 'export NXF_ENABLE_SECRETS=true' >> ~/.bashrc
source ~/.bashrc
nextflow secrets put -n SYNAPSE_AUTH_TOKEN -v "<synapse-pat>"
mkdir -p $HOME/.nextflow/
echo 'aws.client.anonymous = true' >> $HOME/.nextflow/config

# Download and extract Ensembl VEP cache
mkdir -p $HOME/ref/ $HOME/.vep/
rsync -avr --progress rsync://ftp.ensembl.org/ensembl/pub/release-107/variation/indexed_vep_cache/homo_sapiens_vep_107_GRCh38.tar.gz $HOME/ref/
tar -zvxf $HOME/ref/homo_sapiens_vep_107_GRCh38.tar.gz -C $HOME/.vep/

# Download reference genome FASTA file
mkdir -p $HOME/ref/fasta/
aws --no-sign-request s3 sync s3://sage-igenomes/igenomes/Homo_sapiens/GATK/GRCh38/Sequence/WholeGenomeFasta/ $HOME/ref/fasta/

# Stage reference files in memory
mkdir -p /dev/shm/vep/ /dev/shm/fasta/
sudo mount -o remount,size=25G /dev/shm  # Increase /dev/shm size
rsync -avhP $HOME/.vep/ /dev/shm/vep/
rsync -avhP $HOME/ref/fasta/ /dev/shm/fasta/

Preparing the Ensembl VEP cache

Outside of Nextflow

To determine the most efficient way of preparing the VEP cache for vcf2maf, I tried different permutations of downloading the tarball or extracted folder from Ensembl or S3. Here are the individual results:

• Download tarball using rsync from Ensembl: 10 min 23 sec
• Download tarball using AWS CLI from S3: 3 min 14 sec
• Extract tarball using tar locally: 6 min 11 sec
• Download extracted folder using AWS CLI from S3: 4 min 5 sec

Based on the above results, here are some estimated runtimes:

• Download tarball from Ensembl and extract locally: 16 min 34 sec
• Download tarball from S3 and extract locally: 9 min 25 sec
• Download extracted tarball from S3: 4 min 5 sec

Based on the above estimates, downloading the extracted tarball from S3 seems like the most efficient method followed by downloading the tarball from S3 and extracting locally.

Within Nextflow

After benchmarking different methods for downloading VEP cache, I performed various test within a Nextflow run. Note that SHM refers to files being in shared memory (i.e. /dev/shm).

• Baseline (all reference files in SHM): 3 min 43 sec
• FASTA in S3 and VEP folder in SHM: 4 min 9 sec
• FASTA in S3 and VEP folder in non-SHM: 3 min 43 sec
• FASTA in S3 and VEP folder in S3: Over 17 min 7 sec¹
• FASTA in S3 and VEP tarball in non-SHM: 8 min 39 sec
• FASTA and VEP tarball in S3: 8 min 38 sec

The above results showed that downloading the tarball from S3 was the most efficient method. While ~10 minutes is a long time to spend on preparing reference files, it's trivial compared to the actual runtimes of vcf2maf, which can reach 4-5 hours. The benefit is portability, including the ability of running this workflow on Tower.

Footnotes

1. While this was expected to be the most efficient method of downloading the VEP cache, I had to kill the job because it was taking so long. Perhaps the AWS Java SDK isn't as efficient as the AWS CLI for downloading a folder in S3 recursively. ↩