vcf-js

High performance Variant Call Format (VCF) parser in pure JavaScript.

Status

Usage

This module is best used when combined with some easy way of retrieving the header and individual lines from a VCF, like the @gmod/tabix module.

import { TabixIndexedFile } from '@gmod/tabix'

// with import
import VCF, { parseBreakend } from '@gmod/vcf'

// with require
const { default: VCF, parseBreakend } = require('@gmod/vcf')

const tbiIndexed = new TabixIndexedFile({ path: '/path/to/my.vcf.gz' })

async function doStuff() {
  const headerText = await tbiIndexed.getHeader()
  const tbiVCFParser = new VCF({ header: headerText })
  const variants = []
  await tbiIndexed.getLines('ctgA', 200, 300, line =>
    variants.push(tbiVCFParser.parseLine(line)),
  )
  console.log(variants)
}

If you want to stream a VCF file, you can alternatively use something like this

const fs = require('fs')
const VCF = require('@gmod/vcf').default
const { createGunzip } = require('zlib')
const readline = require('readline')

const rl = readline.createInterface({
  input: fs.createReadStream(process.argv[2]).pipe(createGunzip()),
})

let header = []
let elts = []
let parser = undefined

rl.on('line', function (line) {
  if (line.startsWith('#')) {
    header.push(line)
    return
  } else if (!parser) {
    parser = new VCF({ header: header.join('\n') })
  }
  const elt = parser.parseLine(line)
  elts.push(elt.INFO.AN[0])
})

rl.on('close', function () {
  console.log(elts.reduce((a, b) => a + b, 0) / elts.length)
})

This method is used to test @gmod/vcf in https://github.com/brentp/vcf-bench

Methods

Given a VCF with a single variant line

#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	HG00096
contigA	3000	rs17883296	G	T,A	100	PASS	NS=3;DP=14;AF=0.5;DB;XYZ=5	GT:AP	0|0:0.000,0.000

The variant object returned by parseLine() would be

{
  CHROM: 'contigA',
  POS: 3000,
  ID: ['rs17883296'],
  REF: 'G',
  ALT: ['T', 'A'],
  QUAL: 100,
  FILTER: 'PASS',
  INFO: {
    NS: [3],
    DP: [14],
    AF: [0.5],
    DB: true,
    XYZ: ['5'],
  },
  SAMPLES: () => ({
    HG00096: {
      GT: ['0|0'],
      AP: ['0.000', '0.000'],
    },
  }),
  GENOTYPES: () => ({
    HG00096: '0|0',
  }),
}

The variant.SAMPLES() and variant.GENOTYPES() are functions because it does not try to eagerly parse all the genotype data, so will only do so when you call either of these which can save time especially if your VCF has a lot of samples in it.

The variant.SAMPLES() function parses out the FORMAT fields, while variant.GENOTYPES() returns just the genotypes string which can be faster if that is the only information you are interested in

The parser will try to convert the values in INFO and FORMAT to the proper types using the header metadata. For example, if there is a header line like

##INFO=<ID=ABC,Number=2,Type=Integer,Description="A description">

The parser will expect any INFO entry ABC to be an array of two integers, so it would convert ABC=12,20 to { ABC: [12, 20] }.

Each INFO entry value will be an array unless Type=Flag is specified, in which case it will be true. If no metadata can be found for the entry, it will assume Number=1 and Type=String.

Some fields are pre-defined by the VCF spec, which is why in the variant object above "DP" was parsed as an integer (it is defined in the VCF spec), but "XYZ" was left as a string (it is not defined in either the VCF spec or the header).

Metadata can be accessed with the getMetadata() method, including all the built-in metadata from the VCF spec. With no parameters it will return all the data. Any parameters passed will further filter the metadata. For example, for a VCF with this header:

##INFO=<ID=ABC,Number=2,Type=Integer,Description="A description">
#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO

you can access the VCF's header metadata like (some output omitted for clarity):

> console.log(vcfParser.getMetadata())
{ INFO:
   { AA:
      { Number: 1, Type: 'String', Description: 'Ancestral Allele' },

...

     ABC: { Number: 2, Type: 'Integer', Description: 'A description' } },
  FORMAT:
   { AD:
      { Number: 'R',
        Type: 'Integer',
        Description: 'Read depth for each allele' },

...

  ALT:
   { DEL: { Description: 'Deletion relative to the reference' },

...

  FILTER: { PASS: { Description: 'Passed all filters' } } }

> console.log(vcfParser.getMetadata('INFO'))
{ AA:
   { Number: 1, Type: 'String', Description: 'Ancestral Allele' },
  AC:
   { Number: 'A',
     Type: 'Integer',
     Description:
      'Allele count in genotypes, for each ALT allele, in the same order as listed' },
  AD:
   { Number: 'R',
     Type: 'Integer',
     Description: 'Total read depth for each allele' },

...

  ABC: { Number: 2, Type: 'Integer', Description: 'A description' } }

> console.log(vcfParser.getMetadata('INFO', 'DP'))
{ Number: 1, Type: 'Integer', Description: 'Total Depth' }

> console.log(vcfParser.getMetadata('INFO', 'DP', 'Number'))
1

A list of sample names is also available in the samples attribute of the parser object:

> console.log(vcfParser.samples)
[ 'HG00096' ]

Breakends

We offer a helper function to parse breakend strings. We used to parse these automatically but it is now a helper function

import { parseBreakend } from '@gmod/vcf'
parseBreakend('C[2:321682[')

// output
//
//     {
//       "MateDirection": "right",
//       "Replacement": "C",
//       "MatePosition": "2:321682",
//       "Join": "right"
//     }

• The C[2:321682[ parses as "Join": "right" because the BND is after the C base
• The C[2:321682[ also is given "MateDirection": "right" because the square brackets point to the right.
• The spec never has the square brackets pointing in different directions. Instead, the different types of joins can be imagined as follows

For the above vcf line where chr13:123456->C[2:321682[ then we have this

    chr13:123456
  -------------C\
                 \
                  \
                   \
                    \
                     \
                      \
                       \
                        \--------------
                         chr2:321682

If the alt was instead chr13:123456->[2:321682[C then the the "Join" would be "left" since the "BND" is before "C" and then the breakend structure looks like this

      chr13:123456

      |C--------------------
      |
      |
      |
      |
      |
      |
      |
      |
      |
      |
      ----------------------
       chr2:321682

API

Table of Contents

• VCFParser
  • Parameters
  • getMetadata
    • Parameters
  • parseLine
    • Parameters

VCFParser

Class representing a VCF parser, instantiated with the VCF header.

Parameters

• args object

  • args.header string The VCF header. Supports both LF and CRLF newlines. (optional, default '')
  • args.strict boolean Whether to parse in strict mode or not (default true) (optional, default true)

getMetadata

Get metadata filtered by the elements in args. For example, can pass ('INFO', 'DP') to only get info on an metadata tag that was like "##INFO=<ID=DP,...>"

Parameters

• args ...string List of metadata filter strings.

Returns any An object, string, or number, depending on the filtering

parseLine

Parse a VCF line into an object like

{
  CHROM: 'contigA',
  POS: 3000,
  ID: ['rs17883296'],
  REF: 'G',
  ALT: ['T', 'A'],
  QUAL: 100,
  FILTER: 'PASS',
  INFO: {
    NS: [3],
    DP: [14],
    AF: [0.5],
    DB: true,
    XYZ: ['5'],
  },
  SAMPLES: () => ({
    HG00096: {
      GT: ['0|0'],
      AP: ['0.000', '0.000'],
    }
  }),
  GENOTYPES: () => ({
    HG00096: '0|0'
  })
}

SAMPLES and GENOTYPES methods are functions instead of static data fields because it avoids parsing the potentially long list of samples from e.g. 1000 genotypes data unless requested.

The SAMPLES function gives all info about the samples

The GENOTYPES function only extracts the raw GT string if it exists, for potentially optimized parsing by programs that need it

Parameters

• line string A string of a line from a VCF