This year's Annovar tutorial has been split into 2 separate tutorials. The the basic introduction to the tool and working with output files that have previously been generated is presented here.

The combined tutorial that includes steps for running the analysis yourself can now be found here

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.


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. You may be asking yourself "where can I find this awesome program?", but hopefully by now your assumption is that it is either on TACC or in the BioITeam folder. Generally speaking "programs" that consist of a series of scripts without many complex dependencies can easily be installed in the $BI folders. While the most popular programs will eventually be turned into modules. Despite its power, you can find this program in the $BI folders.

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.


Accessing pre-computed results

Use this code block to copy the completed results to immediately begin evaluating the results. Additionally, when your results have completed running (likely >90 minutes as annovar not configured to use multiple processes), compare your results to these provided results.
mkdir $SCRATCH/GVA_Annovar_provided_results
cd $SCRATCH/GVA_Annovar_provided_results
cp $BI/ngs_course/human_variation/*chrom20.samtools* .
cp $BI/ngs_course/human_variation/*chrom20.GATK* .


ANNOVAR output

Two sets of output are provided. One for samtools and another for GATK with simple substitutions in the name for each of 3 individuals:

Example ANNOVAR output on the NA12878 vcf file
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

The exome_summary.csv is probably 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 thAnnovar filter descriptions page here):


Funcexonic, splicing, ncRNA, UTR5, UTR3, intronic, upstream, downstream, intergenic
GeneThe common gene name
ExonicFuncframeshift insertion/deletion/block subst, stopgain, stoploss, nonframeshift ins/del/block stubst., nonsynonymous SNV, synonymous SNV, or Unknown
AAChange (in gene coordinates)
Conserved (i.e. SNP is in a conserved region)based on the UCSC 46-way conservation model
SegDup (snp is in a segmental dup. region)
ESP5400_ALLAlternate Allele Frequency in 3510 NHLBI ESP European American Samples
1000g2010nov_ALL

Alternative Allele Frequency in 1000 genomes pilot project 2012 Feb release (minor allele could be reference or alternative allele).

dbSNP132The id# in dbSNP if it exists
AVSIFTThe AVSIFT score of how deleterious the variant might be
LJB_PhyloPConservation score provided by dbNSFP which is re-scaled from original phylop score. The new score ranges from 0-1 with larger scores signifying higher conservation. A recommended cutoff threshold is 0.95. If the score > 0.95, the prediction is "conservative". if the score <0.95, the prediction is "non-conservative". 
LJB_PhyloP_Pred
LJB_SIFTSIFT takes a query sequence and uses multiple alignment information to predict tolerated and deleterious substitutions for every position of the query sequence.  Positions with normalized probabilities less than 0.05 are predicted to be deleterious, those greater than or equal to 0.05 are predicted to be tolerated. 
LJB_SIFT_Pred
LJB_PolyPhen2

Functional prediction score for non-syn variants from Polyphen2 provided by dbNSFP  (higher score represents functionally more deleterious). A score greater than 0.85 corresponds to prediciton of "probably damaging". The prediciton is "possbily damaging" if score is between 0.85 and 0.15, and "benign" if score is below 0.15.

LJB_PolyPhen2_Pred
LJB_LRTFunctional prediction score for non-syn variants from LRT provided by dbNSFP (higher score represents functionally more deleterious. It ranges from 0 to 1. This score needs to be combined with other information prediction. If a threshold has to be picked up under some situation, 0.995 can be used as starting point. 
LJB_LRT_Pred
LRT_MutationTaster

Functional prediction score for non-syn variants from Mutation Taster provided by dbNSFP  (higher score represents functionally more deleterious). The score ranges from 0 to 1. Similar to LRT, the prediction is not entirely depending on the score alone. But if a threshold has to be picked, 0.5 is the recommended as the starting point.  

LRT_MutationTaster_Pred
LJB_GERP++higher scores are more deleterious
Chr
Start
End
RefReference base
ObsObserved base-pair or variant
SNP Quality value
filter information
(ALL the VCF info is here!!)
GT:PL:GQ for each file!


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!

The final answer is "DEFB126"

grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv  # this will print all the lines which contains the text "frameshift"
# From the output you can key into the first few columns having the information you are interested in: location-classification, gene, mutation type each separated by commas. This should lead you to think about adding the awk command to print only some columns.
 
grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}'  # the -F"," syntax forces it to split on commas
# you will likely notice this data is easier to visualize, and in this case you can probably see what gene is represented multiple times, but why stop there ... lets add the uniq -c command to the pipes to have linux count for us
 
grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}' | uniq -c 
# for the number of mutations we have this is sufficient, but for increased numbers of mutations where you may be interested in displaying them in a particular order. This can be done by adding the sort command

grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}' | uniq -c | sort -r  # the -r option on the sort command  sorts in reverse order
grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}' | uniq -c | sort -r
grep "frameshift" NA12878.chrom20.samtools.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}' | uniq -c | sort -r

GATK detects 2 frameshift insertions in DNMT3B, and 1 each in PRNP and LAMA5. SAMtools detects 1 frameshift deletion in each of NTSR1 and CTSA that GATK does not detect.

Compare the output of these three commands:

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

Do you notice a pattern?

What's the right statistical test to determine whether non-synonymous mutations might be under different selective pressure than intergenic or synonymous mutations from this data?

The final form of the command that you had was:
grep "frameshift" NA12878.chrom20.GATK.vcf.exome_summary.csv | awk -F"," '{print $2"\t"$3}' | uniq -c | sort -r
Try to come up with a solution on your own before checking your answer here
cat *GATK.vcf.exome_summary.csv | grep "exonic" | awk -F"," '{print $2,"\t"$3}' | sort | uniq -c | sort -nr

Some questions for thought:

  1. Why might this type of analysis be useful?
  2. What other greps would make this more useful?


Other variant annotators:

Return to GVA2021


  • No labels