Annotating Variants: Introduction

As we've already seen, determining the presence or absence of a variant from NGS data is not trivial. It is software dependent and has inherent trade-offs between sensitivity and specificity. Inevitably, the number of putative variants in a real data set is very large; for example the first samples from the 1000 Genomes project typically found 0.1% variants (3 million variants), approximately 10% of which had never been previously observed (300,000 novel variants per individual.) False positive discovery rates are also typically very high at this stage.

Auxiliary data is often used to reduce putative variants without compromising sensitivity. Examples of auxiliary data include other samples within a cohort, existing SNP databases, gene or other feature annotations, and sample-specific information such as pedigree:

• By comparing genotypes across a set of samples and defining one as "reference" (or "wild type") enables other samples to be properly genotyped (i.e. 0/0 for hom. WT, 0/1 for het, 1/1 for hom. alt)
• Existing SNP databases such as dbSNP or the vcf files from the 1000 genomes project may be used to reject "common" variants under the assertion that "common" means "non-disease causing".
• Gene annotations allow for codon analysis to determine whether mutations are synonymous, non-synonymous, nonsense, mis-sense, or create early stop codons.
• Pedigree information is particularly effective in mendelian autosomal recessive diseases; filtering for heterozygous mutations in parents and unaffected siblings which are homozygous in the proband usually yields a very small set of candidate variants.

Variant annotation tools perform the function of combining the raw putative variant calls with auxiliary data to add meaning ("annotation") to the variants. In many cases, the variant detection tool itself will add certain elements of annotation, such as a definition of the variant, a genotype call, a measure of likelihood, a haplotype score, and other measures of the raw data useful to reduce false positives. In other cases, the annotator will only require a vcf file combined with other auxiliary data.

Because these tools draw in information from may disparate sources, they can be very difficult to install, configure, use, and maintain. For example, the vcf files from the 1000 Genomes project are arranged in a deep ftp tree by date of data generation. Large genome centers spend significant resources managing these tools. Our objective

Pre-packaged programs

Annovar - one of the most powerful yet simple to run variant annotators available

Annovar is a variant annotator. Given a vcf file from an unknown sample and a host of existing data about genes, other known SNPs, gene variants, etc., Annovar will place the discovered variants in context.

Annovar comes pre-packaged with human auxiliary data which is updated by the authors on a regular basis. It is a well-constructed package in that there is one core program annotate_variation.pl which can perform a variety of different types of annotation AND download the reference databases required.

The authors have also included a wrapper script summarize_anovar.pl which runs a fairly comprehensive set of annotations automatically.

This next exercise will give you some idea of how Annovar works; we've taken the liberty of writing the bash script annovar_pipe.sh around the existing summarize_annovar.pl wrapper (a wrapper within a wrapper - a common trick) to even further simplify the process for this course.

Exercise:

First, look at the code for our annovar_pipe.sh command. Here is an easy one-liner to cat the contents of a script (note ` is a back-tick, not apostrophe):

cat `which annovar_pipe.sh`

This script simply does a format conversion and then calls summarize_annovar.pl. Now let's run it on all the vcf files - you could simply edit the commands file and type in the 6 lines, or you can use this fancier command line that calls Perl to custom-create the 6 command lines needed and put them straight into commands:

ls $BI/ngs_course/human_variation/N*.vcf | \
  perl -n -e 'chomp; $_=~/(NA\d+).*(sam|GATK)/; print "annovar_pipe.sh $_ >$1.$2.log 2>&1\n";' \
  > commands

launcher_creator.py -l annovar.sge -n annovar -t 00:30:00 -j commands
qsub annovar.sge

We have ALREADY pre-computed these outputs (although Annovar will run pretty quickly on data from only chr20).  You might want to have a look at the code to annovar_pipe.sh and summarize_annovar.pl.  Note that these run Annovar in "gene-based" mode.

more `which annovar_pipe.sh`

more `which summarize_annovar.pl`

ANNOVAR output

Annovar does a ton of work in assessing variants for us (though if you were going for clinical interpretation, you still have a long way to go - compare this to RUNES or CarpeNovo).  It provides all these output files:

NA12878.chrom20.samtools.vcf.exome_summary.csv
NA12878.chrom20.samtools.vcf.exonic_variant_function
NA12878.chrom20.samtools.vcf.genome_summary.csv
NA12878.chrom20.samtools.vcf.hg19_ALL.sites.2010_11_dropped
NA12878.chrom20.samtools.vcf.hg19_ALL.sites.2010_11_filtered
NA12878.chrom20.samtools.vcf.hg19_avsift_dropped
NA12878.chrom20.samtools.vcf.hg19_avsift_filtered
NA12878.chrom20.samtools.vcf.hg19_esp5400_all_dropped
NA12878.chrom20.samtools.vcf.hg19_esp5400_all_filtered
NA12878.chrom20.samtools.vcf.hg19_genomicSuperDups
NA12878.chrom20.samtools.vcf.hg19_ljb_all_dropped
NA12878.chrom20.samtools.vcf.hg19_ljb_all_filtered
NA12878.chrom20.samtools.vcf.hg19_phastConsElements46way
NA12878.chrom20.samtools.vcf.hg19_snp132_dropped
NA12878.chrom20.samtools.vcf.hg19_snp132_filtered
NA12878.chrom20.samtools.vcf.log
NA12878.chrom20.samtools.vcf.variant_function

I find the exome_summary.csv to be one of the most useful files because it brings together nearly all the useful information.  Here are the fields in that file (see these docs for more information, or the Annovar filter descriptions page here):

Everything after the "LJB_GERP++" field in exome_summary came from the original VCF file, so this file REALLY contains everything you need to go on to functional analysis!  This is one of the many reasons I like Annovar.

Scavenger hunts!

Find the gene with two frameshift deletions in NA12878

Test "genetic drift" vs. "functional selection" - e.g. is the distribution of variants different among non-coding regions, synonymous changes in coding regions, and non-synonymous changes in coding regions?

grep intergenic NA12878.chrom20.samtools.vcf.genome_summary.csv | awk 'BEGIN {FS=","} {print $26"\t"$27}' | sort | uniq -c | sort -n -r | head -20
grep exonic NA12878.chrom20.samtools.vcf.genome_summary.csv | grep -w synonymous | awk 'BEGIN {FS=","} {print $25"\t"$26}' | sort | uniq -c | sort -n -r | head -20
grep exonic NA12878.chrom20.samtools.vcf.genome_summary.csv | grep -w nonsynonymous | awk 'BEGIN {FS=","} {print $25"\t"$26}' | sort | uniq -c | sort -n -r | head -20

Other variant annotators:

• http://www.yandell-lab.org/software/vaast.html
• http://www.broadinstitute.org/gatk/gatkdocs/ VariantAnnotator annotations
• http://www.bioconductor.org/help/workflows/variants/
• http://vat.gersteinlab.org/
• http://code.google.com/p/mu2a/

Notes

Variants consist of single base base changes, insertions and deletions, and larger scale structural changes. "Larger scale" is usually defined relative to the capabilities of the technology; for example, a "small indel" usually means "detectable within a single sequence read". In 2009, sequence reads were about 50 bp but in 2011 they were 100 bp.