This repo provides tools to convert ClinVar data into a tab-delimited flat file, and also provides that resulting tab-delimited flat file.
The full set of tables generated by the pipeline for the latest clinvar release is stored in the output folder.
These tables are separated by genome build (output/b37 or output/b38) and whether they represent
simple mono-allelic variants (../single/
sub-directory),
or more complex variants such as compound-heterozygotes, haplotypes, etc. (../multi/
sub-directory) - for a total of 4 directories that contain tables.
Each of these directories contains the following files:
- clinvar_allele_trait_pairs.*.tsv.gz: table where each row represents an allele-trait pair.
- clinvar_alleles.*.tsv.gz: table where each row represents a single variant allele. This is generated by grouping clinvar_allele_trait_pairs.tsv.gz by allele.
- clinvar_alleles.*.vcf.gz: clinvar_alleles.tsv.gz converted to VCF format.
- clinvar_alleles_with_exac_v1.*.tsv.gz: clinvar_alleles.tsv.*.gz with additonal columns from the ExAC v1 dataset that have non-empty values for all clinvar alleles that are also present in ExAC.
- clinvar_alleles_with_gnomad_exomes.*.tsv.gz: clinvar_alleles.tsv.*.gz with additonal columns from the gnomAD dataset that have non-empty values for all clinvar alleles that are also present in the gnomAD exomes callset.
- clinvar_alleles_with_gnomad_genomes.*.tsv.gz: clinvar_alleles.tsv.*.gz with additonal columns from the gnomAD dataset that have non-empty values for all clinvar alleles that are also present in the gnomAD genomes callset.
- clinvar_alleles_stats.*.txt: summary of the different columns in clinvar_alleles.*.tsv.gz, along with some basic stats on the different values that appear in each column.
ClinVar is a public database hosted by NCBI for the purpose of collecting assertions as to genotype-phenotype pairings in the human genome. One common use case for ClinVar is as a catalogue of genetic variants that have been reported to cause Mendelian disease. In our work in the MacArthur Lab, we have two major use cases for ClinVar:
- To check whether candidate causal variants we find in Mendelian disease exomes have been previously reported as pathogenic.
- To pair with ExAC data to enable exome-wide analyses of reportedly pathogenic variants.
ClinVar makes its data available via FTP in three formats: XML, TXT, and VCF. We found that none of these files were ideally suited for our purposes. The VCF only contains variants present in dbSNP; it is not a comprehensive catalogue of ClinVar variants. The TXT file lacks certain annotations such as PubMed IDs for related publications. The XML file is large and complex, with multiple entries for the same genomic variant, making it difficult to quickly look up a variant of interest. In addition, both the XML and TXT representations are not guaranteed to be unique on genomic coordinates, and also contain many genomic coordinates that have been parsed from HGVS notation, and therefore may be right-aligned (in contrast to left alignment, the standard for VCF) and may also be non-minimal (containing additional nucleotides of context to the left or right of a given variant).
To create a flat representation of ClinVar suited for our purposes, we took several steps, encapsulated in the pipeline src/master.py
- Download the latest XML and TXT dumps from ClinVar FTP.
- Parse the XML file using src/parse_clinvar_xml.py to extract fields of interest into a flat file.
- Normalize using our Python implementation of vt normalize (see [Tan 2015]).
- Group the allele-trait records by allele using src/group_by_allele.py to aggregate interpretations from multiple submitters by allele, independent of conditions.
- Join the TXT file using src/join_data.R to aggregate interpretations from multiple submitters independent of conditions.
- Generate the VCF file and other tables based on the file created in 5.
†Because a ClinVar record may contain multiple assertions of Clinical Significance, we defined three additional columns in encoding the Clinical Significance on ClinVar aggregated records:
pathogenic
is1
if the variant is reported as "Pathogenic" or "Likely Pathogenic" without any "uncertain significance" submissionbenign
is1
if the variant is reported as "Benign" or "Likely benign" without any "uncertain significance" submissionconflicted
is1
if the variant is reported as "Conflicting interpretation of pathogenicity" or "conflicting data from submitters".
The pipeline scripts expect the following programs to be available on your system (and in your $PATH
):
wget
python2.7
Rscript
tabix
vt
To run the pipeline:
cd ./src
pip install --user --upgrade -r requirements.txt
python2.7 master.py --b37-genome /path/to/b37.fa --b38-genome /path/to/b38.fa -E /path/to/ExAC.r1.sites.vep.vcf.gz -GG /path/to/gnomad.genomes.r2.0.1.sites.coding.autosomes_and_X.vcf.gz -GE /path/to/gnomad.exomes.r2.0.1.sites.vcf.gz
See python master.py -h
for additional options.
Because ClinVar contains a great deal of data complexity, we made a deliberate decision to not attempt to capture all fields in our resulting file. We made an effort to capture a subset of fields that we believed would be most useful for genome-wide filtering, and also included measureset_id
as a column to enable the user to look up additional details on the ClinVar website. For instance, the page for the variant with measureset_id
7105 is located at ncbi.nlm.nih.gov/clinvar/variation/7105/. Note that we also do not capture all of the complexity of the fields that are included.
If you want to analyze the output file in R, a suitable line of code to read it in would be:
clinvar = read.table('clinvar_alleles.tsv',sep='\t',header=T,quote='',comment.char='')
The accuracy of output files is limited by the download-able files from the ClinVar FTP site. In the case that ClinVar releases new data with a new reporting format or an unfinished format update, our pipeline may not work for the latest release. We would recommend that users revert to the old version by specifying the input ClinVar files when executing python master.py
.
ClinVar data, as a work of the United States federal government, are in the public domain and are redistributed here under the same terms as they are distributed by ClinVar itself. Importantly, note that ClinVar data are "not intended for direct diagnostic use or medical decision-making without review by a genetics professional". The code in this repository is distributed under an MIT license.
Zhang X, Minikel EV, O'Donnell-Luria AH et al. ClinVar data parsing [version 1; referees: 2 approved]. Wellcome Open Res 2017, 2:33 (doi: 10.12688/wellcomeopenres.11640.1)