vcfanno annotates a VCF with any number of sorted and tabixed input BED, BAM, and VCF files in parallel. It does this by finding overlaps as it streams over the data and applying user-defined operations on the overlapping annotations.

In order to parallelize, work is broken down as follows. A slice (array) of query intervals is accumulated until a specified number is reached (usually ~5K-25K) or a gap cutoff is exceeded; at that point, the bounds of the region are used to perform a tabix (or any regional) query on the database files. This is all done in irelate. vcfanno then iterates over the streams that result from the tabix queries and finds intersections with the query stream. This is a parallel chrom-sweep. This method avoids problems with chromosome order.

For VCF, values are pulled by name from the INFO field. For BED, values are pulled from (1-based) column number. For BAM, depth (count), "mapq" and "seq" are currently supported.

vcfanno is written in go It can annotate more than 8,000 variants per second with 34 annotations from 9 files on a modest laptop.

We are actively developing vcfanno and appreciate feedback and bug reports.

After downloading the binary for your system (see section below) usage looks like:

  ./vcfanno -js example/custom.js example/conf.toml example/query.vcf.gz

Where conf.toml looks like:

[[annotation]]
file="ExAC.vcf"
fields = ["AC_AFR", "AC_AMR", "AC_EAS"]
ops=["first", "first", "min"]

[[annotation]]
file="fitcons.bed"
columns = [4, 4]
names=["fitcons_mean", "js_sum"]
# note the 2nd op here is javascript that has access to `vals`
ops=["mean", "js:sum=0;for(i=0;i<vals.length;i++){sum+=vals[i]}; vals"]

[[annotation]]
file="example/ex.bam"
names=["ex_bam_depth"]
fields=["depth", "mapq", "seq"]
ops=["count", "mean", "concat"]

So from ExAC.vcf we will pull the fields from the info field and apply the corresponding operation from the ops array. Users can add as many [[annotation]] blocks to the conf file as desired. Files can be local as above, or available via http/https.

Also see the additional usage section at the bottom for additional details.

the example directory contains the data and conf for a full example. To run, either download the appropriate binary for your system or build with:

go get
go build -o vcfanno

from this directory. Then, you can annotate with:

GOMAXPROCS=4 ./vcfanno -js example/custom.js example/conf.toml example/query.vcf.gz > annotated.vcf

An example INFO field row before annotation (pos 98683):

AB=0.282443;ABP=56.8661;AC=11;AF=0.34375;AN=32;AO=45;CIGAR=1X;TYPE=snp

and after:

AB=0.2824;ABP=56.8661;AC=11;AF=0.3438;AN=32;AO=45;CIGAR=1X;TYPE=snp;AC_AFR=0;AC_AMR=0;AC_EAS=0;fitcons_mean=0.061;js_sum=0.061

In most cases, we will have a single annotation entry for each entry (variant) in the query VCF. However, it is possible that there will be multiple annotations from a single annotation file--in this case, the op determines how the many values are reduced. Valid operations are:

• js:$javascript // see section below for more details
• mean
• max
• min
• concat // comma delimited list of output
• count // count the number of overlaps
• uniq
• first
• flag // presense/absence via vcf flag

binary executables are available here for linux, mac (darwin), and windows for 32 and 64 bit platforms.

Annotations will be the most accurate if your query and annotation variants are split (no multiple ALTs) and normalize (left-aligned and trimmed). At some point, this will be done internally, but for now, you can get a split and normalized VCF using vt with:

vt decompose -s $VCF | vt normalize -r $REF - > $NORM_VCF

This, and the associated go libraries (vcfgo, irelate, xopen) are under active development. The following are on our radar (most have been completed):

• allow annotating with bam fields, e.g. QUAL and SEQ.
• decompose, normalize, and get allelic primitives for variants on the fly (we have code to do this, it just needs to be integrated)
• allow custom golang ops when using api.
• improve test coverage for vcfanno (still need more tests for bam)
• embed otto js engine to allow custom ops.
• support for annotating BED files.

For annotating large variants, such as CNVs or structural variants (SVs), it can be useful to annotate the ends of the variant in addition to the region itself. To do this, specify the -ends flag to vcfanno. e.g.:

vcfanno -ends example/conf.toml example/query.vcf.gz

In this case, the names field in the conf file contains, "fitcons_mean". The output will contain fitcons\_mean as before along with left\_fitcons\_mean and right\_fitcons\_mean for any variants that are longer than 1 base. The left end will be for the single-base at the lowest base of the variant and the right end will be for the single base at the higher numbered base of the variant.

By default, when annotating with a variant, in addition to the overlap requirement, the variants must share the same position, the same reference allele and at least one alternate allele (this is only used for variants, not for BED/BAM annotations). If this flag is specified, only overlap testing is used and shared REF/ALT are not required.

Set to the number of processes that vcfanno can use during annotation. vcfanno parallelizes well up to 15 or so cores.

custom in ops (javascript). For use when the built-in ops don't supply the needed reduction.

we embed the javascript engine otto so that it's possible to create a custom op if it is not provided. For example if the users wants to

"js:sum=0;for(i=0;i<vals.length;i++){sum+=vals[i]};sum"

where the last value (in this case sum) is returned as the annotation value. It is encouraged to instead define javascript functions in separate .js file and point to it when calling vcfanno using the -js flag. So, in an external file, "some.js", instead put:

function sum(vals) {
 	s = 0;
	for(i=0; i<vals.length; i++){
		s+=vals[i]
	}
	return s
}

And then the above custom op would be: "js:sum(vals)". (note that there's a sum op provided by vcfanno which will be faster).

The variables vals, chrom, start, end from the current variant will all be available in the javascript code.

See example/conf.toml and example/custom.js for more examples.