To convert a VCF into a MAF, each variant must be mapped to only one of all possible gene transcripts/isoforms that it might affect. But even within a single isoform, a Missense_Mutation
close enough to a Splice_Site
, can be labeled as either in MAF format, but not as both. This selection of a single effect per variant, is often subjective. And that's what this project attempts to standardize. The vcf2maf
and maf2maf
scripts leave most of that responsibility to Ensembl's VEP, but allows you to override their "canonical" isoforms, or use a custom ExAC VCF for annotation. Though the most useful feature is the extensive support in parsing a wide range of crappy MAF-like or VCF-like formats we've seen out in the wild.
Find the latest release, download it, and view the detailed usage manuals for vcf2maf
and maf2maf
:
export VCF2MAF_URL=`curl -sL https://github.com/gitapi/repos/mskcc/vcf2maf/releases | grep -m1 tarball_url | cut -d\" -f4`
curl -L -o mskcc-vcf2maf.tar.gz $VCF2MAF_URL; tar -zxf mskcc-vcf2maf.tar.gz; cd mskcc-vcf2maf-*
perl vcf2maf.pl --man
perl maf2maf.pl --man
If you don't have VEP installed, then follow this gist. Of the many annotators out there, VEP is preferred for its large team of active coders, and its CLIA-compliant HGVS formats. After installing VEP, test out vcf2maf
like this:
perl vcf2maf.pl --input-vcf tests/test.vcf --output-maf tests/test.vep.maf
To fill columns 16 and 17 of the output MAF with tumor/normal sample IDs, and to parse out genotypes and allele counts from matched genotype columns in the VCF, use options --tumor-id
and --normal-id
. Skip option --normal-id
if you didn't have a matched normal:
perl vcf2maf.pl --input-vcf tests/test.vcf --output-maf tests/test.vep.maf --tumor-id WD1309 --normal-id NB1308
VCFs from variant callers like VarScan use hardcoded sample IDs TUMOR/NORMAL to name genotype columns. To have vcf2maf
correctly locate the columns to parse genotypes, while still printing proper sample IDs in the output MAF:
perl vcf2maf.pl --input-vcf tests/test_varscan.vcf --output-maf tests/test_varscan.vep.maf --tumor-id WD1309 --normal-id NB1308 --vcf-tumor-id TUMOR --vcf-normal-id NORMAL
If VEP is installed under /opt/vep
and the VEP cache is under /srv/vep
, there are options available to tell vcf2maf
where to find them:
perl vcf2maf.pl --input-vcf tests/test.vcf --output-maf tests/test.vep.maf --vep-path /opt/vep --vep-data /srv/vep
If you want to skip running VEP and need a minimalist MAF-like file listing data from the input VCF only, then use the --inhibit-vep
option. If your input VCF contains VEP annotation, then vcf2maf
will try to extract it. But be warned that the accuracy of your resulting MAF depends on how VEP was operated upstream. In standard operation, vcf2maf
runs VEP with very specific parameters to make sure everyone produces comparable MAFs. So, it is strongly recommended to avoid --inhibit-vep
unless you know what you're doing.
If you have a MAF or a MAF-like file that you want to reannotate, then use maf2maf
, which simply runs maf2vcf
followed by vcf2maf
:
perl maf2maf.pl --input-maf tests/test.maf --output-maf tests/test.vep.maf
After tests on variant lists from many sources, maf2vcf
and maf2maf
are quite good at dealing with formatting errors or "MAF-like" files. It even supports VCF-style alleles, as long as Start_Position == POS
. But it's OK if the input format is imperfect. Any variants with a reference allele mismatch are kept aside in a separate file for debugging. The bare minimum columns that maf2maf
expects as input are:
Chromosome Start_Position Reference_Allele Tumor_Seq_Allele2 Tumor_Sample_Barcode
1 3599659 C T TCGA-A1-A0SF-01
1 6676836 A AGC TCGA-A1-A0SF-01
1 7886690 G A TCGA-A1-A0SI-01
See data/minimalist_test_maf.tsv
for a sampler. Addition of Tumor_Seq_Allele1
will be used to determine zygosity. Otherwise, it will try to determine zygosity from variant allele fractions, assuming that arguments --tum-vad-col
and --tum-depth-col
are set correctly to the names of columns containing those read counts. Specifying the Matched_Norm_Sample_Barcode
with its respective columns containing read-counts, is also strongly recommended. Columns containing normal allele read counts can be specified using argument --nrm-vad-col
and --nrm-depth-col
.
Assuming you have a recent version of docker, clone the main branch and build an image as follows:
git clone git@github.com:mskcc/vcf2maf.git
cd vcf2maf
docker build -t vcf2maf:main .
docker builder prune -f
Now you run the scripts in docker as follows:
docker run --rm vcf2maf:main perl vcf2maf.pl --help
docker run --rm vcf2maf:main perl maf2maf.pl --help
A small standalone test dataset was created by restricting VEP v112 cache/fasta to chr21 in GRCh38 and hosting that on a private server for download by CI services. We can manually fetch those as follows:
wget -P tests https://data.cyri.ac/Homo_sapiens.GRCh38.dna.chromosome.21.fa.gz
gzip -d tests/Homo_sapiens.GRCh38.dna.chromosome.21.fa.gz
wget -P tests https://data.cyri.ac/homo_sapiens_vep_112_GRCh38_chr21.tar.gz
tar -zxf tests/homo_sapiens_vep_112_GRCh38_chr21.tar.gz -C tests
And the following scripts test the docker image on predefined inputs and compare outputs against expected outputs:
perl tests/vcf2maf.t
perl tests/vcf2vcf.t
perl tests/maf2vcf.t
Apache-2.0 | Apache License, Version 2.0 | https://www.apache.org/licenses/LICENSE-2.0
Cyriac Kandoth. mskcc/vcf2maf: vcf2maf v1.6. (2020). doi:10.5281/zenodo.593251