Selecting Variants based on annotations in a VCF file

Question

[petersam]

I have a VCF file generated with GATK and annotated with SNPeff. There is an annotation field described as

> ##INFO=<ID=ANN,Number=.,Type=String,Description="Functional annotations: 'Allele | Annotation | Annotation_Impact | Gene_Name | Gene_ID | Feature_Type | Feature_ID | Transcript_BioType | Rank | HGVS.c | HGVS.p | cDNA.pos / cDNA.length | CDS.pos / CDS.length |  AA.pos / AA.length | Distance | ERRORS / WARNINGS / INFO'"

An example of a variant:

AC=5;AF=0.200;AN=10;ANN=C|intron_variant|MODIFIER|GAPDH|4214|transcript|GAPDH_transcript|sim4cc|n.1964-49T>C||||||

I have been using GATK SelectVariants to generate variant lists based on sample genotype and quality metrics but would also like to isolate all variants with different Annotation Impact. In the example above, I would want to get all variants described as "MODIFIER" I am not getting the format correct for the JEXL commands. Can someone suggest a command?

Thank you, S

[Mitch Bekritsky]

Just curious--do you want to produce a valid VCF as output, or just find the records of interest?

Answer 1

[Brice Sarver]

The JEXL stuff in the GATK is wonky at best; I avoid it whenever possible. It is much faster to parse your VCF using grep than it is to use SelectVariants in this case. Grep the lines that have MODIFIER and append them to your VCF. You'll probably need to do standard VCF manipulation (sorting, etc.) after the fact. Something like:

grep MODIFIER yourfile.vcf > my_results.vcf

You will, of course, need to modify your regex to avoid the lines with MODIFIER in the header.

[petersam]

Thank you. That was incredibly easy.

Answer 2

[Mitch Bekritsky]

bcftools view is your friend here. I'd suggest reading the documentation on the linked page for more details, but something along the lines of

bcftools view -i "ANN~MODIFIER" -o modifier.vcf -Ov init.vcf

should get you a VCF file with the variants you're looking for.

Answer 3

[Pierre Lindenbaum]

using vcffilterjs and the following script: https://github.com/lindenb/jvarkit/wiki/VCFFilterJS

function accept(v) {
var ann =   v.getAttributeAsList("ANN");
for(var i in ann)
    {
    var tokens = ann[i].split(/\|/);
    if( tokens.length>2 && tokens[2] == "MODIFIER" ) return true;
    }
return false;
}
accept(variant);