Imported from r2ascension/skill_repo (skills/bio-variant-calling-variant-annotation/SKILL.md). Install upstream with npx skills add r2ascension/skill_repo --skill bio-variant-calling-variant-annotation. Copyright stays with the author.

Version Compatibility

Reference examples tested with: bcftools 1.19+, VEP 110+, SnpEff 5.2+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: <tool> --version then <tool> --help to confirm flags
• Note: SnpEff and SnpSift use single-dash -version, not --version

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Variant Annotation

Tool Comparison

Normalization Before Annotation

Variant normalization is mandatory before annotation. The same variant represented differently (e.g., left-aligned vs right-aligned indels, multiallelic vs biallelic) produces different annotations. Always normalize first:

bcftools norm -f reference.fa -m-any input.vcf.gz -Oz -o normalized.vcf.gz

The -m-any flag splits multiallelic records into biallelic, ensuring each allele gets its own annotation. Without this step, downstream tools may annotate only the first alternate allele or produce ambiguous consequence calls.

Transcript Selection Strategy

Transcript choice determines which consequence is reported. This single decision changes whether a variant is classified as missense, synonymous, or intronic.

VEP flags: --mane_select adds MANE Select transcript info to output. --pick selects one consequence per variant using VEP's internal priority ranking (canonical > biotype > length). For clinical applications, using both together reports the MANE Select consequence when available.

Tool Concordance

Concordance between VEP, SnpEff, and ANNOVAR is lower than commonly assumed: only 58.52% agreement on HGVSc notation, 84.04% on HGVSp, and 85.58% on coding impact classification. Loss-of-function discrepancies affect ACMG PVS1 interpretation in 33-44% of cases.

Practical implication: for clinical applications, annotate with multiple tools and compare results. For research, pick one tool and apply it consistently across all samples. VEP with MANE Select is the current community standard for clinical genomics.

Clinical vs Research Annotation

bcftools annotate

Goal: Add or remove INFO/ID annotations from external databases using bcftools.

Approach: Match variants by position and allele against annotation VCF/BED/TAB files, copying specified columns.

"Add rsIDs to my VCF from dbSNP" → Match variant positions against a database and copy identifiers or annotation fields into the VCF.

Add Annotations from Database

bcftools annotate -a dbsnp.vcf.gz -c ID input.vcf.gz -Oz -o annotated.vcf.gz

Annotation Columns (-c)

Add Multiple Annotations

bcftools annotate -a database.vcf.gz -c ID,INFO/AF,INFO/CAF input.vcf.gz -Oz -o annotated.vcf.gz

Add from BED/TAB Files

# BED with 4th column as annotation
bcftools annotate -a regions.bed.gz -c CHROM,FROM,TO,INFO/REGION \
    -h <(echo '##INFO=<ID=REGION,Number=1,Type=String,Description="Region name">') \
    input.vcf.gz -Oz -o annotated.vcf.gz

# Tab file: CHROM POS VALUE
bcftools annotate -a annotations.tab.gz -c CHROM,POS,INFO/SCORE \
    -h <(echo '##INFO=<ID=SCORE,Number=1,Type=Float,Description="Custom score">') \
    input.vcf.gz -Oz -o annotated.vcf.gz

Remove Annotations

bcftools annotate -x INFO/DP,INFO/MQ input.vcf.gz -Oz -o clean.vcf.gz
bcftools annotate -x INFO input.vcf.gz -Oz -o minimal.vcf.gz  # Remove all INFO

Set ID from Fields

bcftools annotate --set-id '%CHROM\_%POS\_%REF\_%ALT' input.vcf.gz -Oz -o with_ids.vcf.gz

bcftools csq

Goal: Predict functional consequences of variants using gene annotations.

Approach: Map variants to GFF3 gene models and classify as synonymous, missense, frameshift, etc.

Simple consequence prediction using GFF annotation.

bcftools csq -f reference.fa -g genes.gff3.gz input.vcf.gz -Oz -o consequences.vcf.gz

Consequence Types

Ensembl VEP

Goal: Annotate variants comprehensively with consequence, impact, pathogenicity scores, and population frequencies.

Approach: Run VEP with offline cache, enabling SIFT, PolyPhen, HGVS, frequency, and plugin-based predictions.

"Annotate my variants with functional consequences" → Predict coding effects, impact severity, and pathogenicity using Ensembl's Variant Effect Predictor.

Installation

conda install -c bioconda ensembl-vep
vep_install -a cf -s homo_sapiens -y GRCh38 --CONVERT

Basic Annotation

vep -i input.vcf -o output.vcf --vcf --cache --offline

Comprehensive Annotation

vep -i input.vcf -o output.vcf \
    --vcf \
    --cache --offline \
    --species homo_sapiens \
    --assembly GRCh38 \
    --everything \
    --fork 4

--everything Enables

• --sift b - SIFT predictions
• --polyphen b - PolyPhen predictions
• --hgvs - HGVS nomenclature
• --symbol - Gene symbols
• --canonical - Canonical transcript
• --af - 1000 Genomes frequencies
• --af_gnomade/g - gnomAD frequencies
• --pubmed - PubMed IDs

Filter by Impact

vep -i input.vcf -o output.vcf --vcf \
    --cache --offline \
    --pick \
    --filter "IMPACT in HIGH,MODERATE"

Plugins

# CADD scores
vep -i input.vcf -o output.vcf --vcf \
    --cache --offline \
    --plugin CADD,whole_genome_SNVs.tsv.gz

# dbNSFP (multiple predictors)
vep -i input.vcf -o output.vcf --vcf \
    --cache --offline \
    --plugin dbNSFP,dbNSFP4.3a.gz,ALL

# Multiple plugins
vep -i input.vcf -o output.vcf --vcf \
    --cache --offline \
    --plugin CADD,cadd.tsv.gz \
    --plugin dbNSFP,dbnsfp.gz,SIFT_score,Polyphen2_HDIV_score \
    --plugin SpliceAI,spliceai.vcf.gz

VEP Output Fields

SnpEff

Goal: Annotate variants with gene effects and impact categories using SnpEff.

Approach: Run SnpEff ann against a genome database, then use SnpSift for database cross-referencing and filtering.

Installation

conda install -c bioconda snpeff
snpEff download GRCh38.105

Memory

SnpEff loads the genome database into memory as Java objects. Set -Xmx larger than the database size on disk -- human genome databases (e.g. GRCh38) expand to 3-4 GB in memory and need at least -Xmx8g. Smaller genomes can use less. Without sufficient heap, SnpEff will hit OutOfMemoryError or thrash on garbage collection.

Basic Annotation

snpEff ann GRCh38.105 input.vcf > output.vcf

With Statistics

snpEff ann -v -stats stats.html -csvStats stats.csv GRCh38.105 input.vcf > output.vcf

Filter by Impact

snpEff ann GRCh38.105 input.vcf | \
    SnpSift filter "(ANN[*].IMPACT = 'HIGH')" > high_impact.vcf

SnpEff Impact Categories

SnpSift Database Annotations

# dbSNP
SnpSift annotate dbsnp.vcf.gz input.vcf > annotated.vcf

# ClinVar
SnpSift annotate clinvar.vcf.gz input.vcf > annotated.vcf

# dbNSFP
SnpSift dbnsfp -db dbNSFP4.3a.txt.gz input.vcf > annotated.vcf

# Chain multiple
snpEff ann GRCh38.105 input.vcf | \
    SnpSift annotate dbsnp.vcf.gz | \
    SnpSift annotate clinvar.vcf.gz > fully_annotated.vcf

SnpSift Filtering

SnpSift filter "(QUAL >= 30) & (DP >= 10)" input.vcf > filtered.vcf
SnpSift filter "(exists CLNSIG) & (CLNSIG has 'Pathogenic')" input.vcf > pathogenic.vcf

ANNOVAR

Goal: Annotate variants with gene, frequency, and pathogenicity databases using ANNOVAR.

Approach: Run table_annovar.pl with multiple protocols (gene, filter, region) against downloaded annotation databases.

Installation

# Download from https://annovar.openbioinformatics.org/ (registration required)
annotate_variation.pl -buildver hg38 -downdb -webfrom annovar refGene humandb/
annotate_variation.pl -buildver hg38 -downdb -webfrom annovar gnomad30_genome humandb/

Table Annotation

table_annovar.pl input.vcf humandb/ \
    -buildver hg38 \
    -out annotated \
    -remove \
    -protocol refGene,gnomad30_genome,clinvar_20230416,dbnsfp42a \
    -operation g,f,f,f \
    -nastring . \
    -vcfinput

Python: Parse Annotated VCF

Goal: Extract and interpret annotation fields from VEP CSQ or SnpEff ANN strings in Python.

Approach: Parse pipe-delimited annotation strings against the header-defined field order, then filter by impact or consequence.

Parse VEP CSQ

from cyvcf2 import VCF

def parse_vep_csq(csq_string, csq_header):
    fields = csq_header.split('|')
    values = csq_string.split('|')
    return dict(zip(fields, values))

vcf = VCF('vep_output.vcf')
csq_header = None
for h in vcf.header_iter():
    if h['HeaderType'] == 'INFO' and h['ID'] == 'CSQ':
        csq_header = h['Description'].split('Format: ')[1].rstrip('"')
        break

for variant in vcf:
    csq = variant.INFO.get('CSQ')
    if csq:
        for transcript in csq.split(','):
            parsed = parse_vep_csq(transcript, csq_header)
            if parsed.get('IMPACT') in ('HIGH', 'MODERATE'):
                print(f"{variant.CHROM}:{variant.POS} {parsed['SYMBOL']} {parsed['Consequence']}")

Parse SnpEff ANN

from cyvcf2 import VCF

def parse_snpeff_ann(ann_string):
    fields = ['Allele', 'Annotation', 'Impact', 'Gene_Name', 'Gene_ID',
              'Feature_Type', 'Feature_ID', 'Transcript_BioType', 'Rank',
              'HGVS_c', 'HGVS_p', 'cDNA_pos', 'CDS_pos', 'Protein_pos', 'Distance']
    values = ann_string.split('|')
    return dict(zip(fields, values[:len(fields)]))

for variant in VCF('snpeff_output.vcf'):
    ann = variant.INFO.get('ANN')
    if ann:
        for transcript in ann.split(','):
            parsed = parse_snpeff_ann(transcript)
            if parsed['Impact'] == 'HIGH':
                print(f"{variant.CHROM}:{variant.POS} {parsed['Gene_Name']} {parsed['Annotation']}")

Complete Annotation Pipeline

Goal: Run a full annotation workflow from normalization through VEP annotation to impact filtering.

Approach: Normalize variants (mandatory -- same variant represented differently gets different annotations), annotate with VEP using MANE Select transcript selection and comprehensive flags, then filter for HIGH/MODERATE impact.

#!/bin/bash
set -euo pipefail

INPUT=$1
REFERENCE=$2
VEP_CACHE=$3
OUTPUT_PREFIX=$4

# Normalize variants (mandatory before annotation)
bcftools norm -f $REFERENCE -m-any $INPUT -Oz -o ${OUTPUT_PREFIX}_norm.vcf.gz
bcftools index ${OUTPUT_PREFIX}_norm.vcf.gz

# VEP annotation with MANE Select transcript selection
# --mane_select: adds MANE Select transcript info
# --pick: selects one consequence per variant (uses MANE Select when available)
# For research (all transcripts): remove --pick, add --per_gene
vep -i ${OUTPUT_PREFIX}_norm.vcf.gz \
    -o ${OUTPUT_PREFIX}_vep.vcf \
    --vcf --cache --offline --dir_cache $VEP_CACHE \
    --assembly GRCh38 --everything --mane_select --pick --fork 4

bgzip ${OUTPUT_PREFIX}_vep.vcf
bcftools index ${OUTPUT_PREFIX}_vep.vcf.gz

# Filter high/moderate impact
bcftools view -i 'INFO/CSQ~"HIGH" || INFO/CSQ~"MODERATE"' \
    ${OUTPUT_PREFIX}_vep.vcf.gz -Oz -o ${OUTPUT_PREFIX}_filtered.vcf.gz

Pathogenicity Predictors

No single predictor is sufficient for clinical classification. ACMG/AMP guidelines (PP3/BP4) require that computational evidence be treated as supporting, not standalone. ClinGen recommends REVEL >= 0.644 for PP3 (supporting pathogenic) and REVEL <= 0.290 for BP4 (supporting benign) with strong evidence thresholds at >= 0.773 and <= 0.183 respectively.

Clinical Significance (ClinVar)

Quick Reference

Related Skills

• variant-calling/variant-normalization - Normalize before annotating (mandatory preprocessing step)
• variant-calling/filtering-best-practices - Filter by annotations and quality metrics
• variant-calling/clinical-interpretation - ACMG classification and clinical reporting
• variant-calling/vcf-basics - Query annotated fields
• variant-calling/vcf-manipulation - Merge and manipulate annotated VCFs
• database-access/entrez-fetch - Download annotation databases (ClinVar, dbSNP)