domains / science / variant-annotation / variant-annotation

variant-annotation

Skill card

combined

The three skills cover the same high-level task but operate at different levels of abstraction and tooling. The first (aipoch) is a high-level wrapper for a Python script that queries ClinVar/dbSNP for clinical significance. The second (FreedomIntelligence) is a practical guide for using command-line tools (bcftools, VEP, SnpEff, ANNOVAR) to add annotations to VCF files. The third (GPTomics) is a deep-dive into the theoretical and practical pitfalls of variant annotation, focusing on reproducibility, transcript selection (MANE vs RefSeq), and normalization issues. No single copy contains the full procedure; the first lacks the CLI tooling details, the second lacks the deep theoretical context and specific clinical reporting standards (MANE), and the third lacks the simple 'how-to' commands for basic annotation. Combining them provides a complete workflow from basic annotation to clinical-grade reproducibility.

Repositories

Other copies

Combined skill

# Variant Annotation

## 1. Preparation and Normalization
Before annotation, ensure your VCF is normalized to avoid discordant results.
```bash
# Split multiallelics and left-align indels
bcftools norm -f reference.fa -m-any input.vcf.gz -Oz -o normalized.vcf.gz
```
**Critical:** Confirm the genome build (GRCh37 vs GRCh38) of your VCF matches your annotation sources (e.g., gnomAD v2 is GRCh37, v3/v4 are GRCh38).

## 2. Choose Annotation Strategy
- **For Clinical Reporting:** Use MANE Select transcripts on GRCh38. Avoid `VEP --pick` as it is non-deterministic for clinical use.
- **For Research:** Ensembl/GENCODE transcripts may be used, but document the transcript set used.

## 3. Execute Annotation
### Option A: Ensembl VEP (Recommended for Comprehensive Annotation)
```bash
vep -i normalized.vcf -o annotated.vcf \
    --vcf \
    --cache --offline \
    --species homo_sapiens \
    --assembly GRCh38 \
    --mane_select \
    --everything \
    --fork 4
```
*Note: `--everything` includes SIFT, PolyPhen, HGVS, and frequencies. Add `--plugin CADD` or `--plugin dbNSFP` for additional predictors.*

### Option B: SnpEff (Fast Batch Annotation)
```bash
snpEff ann GRCh38.105 normalized.vcf > annotated.vcf
# Add dbSNP/ClinVar
SnpSift annotate dbsnp.vcf.gz annotated.vcf > final_annotated.vcf
```

### Option C: Single Variant Query (Python)
If you only need clinical significance for specific variants (rsID/HGVS), use the `VariantAnnotator` Python class from the aipoch skill:
```python
from scripts.main import VariantAnnotator
annotator = VariantAnnotator()
result = annotator.query_variant("rs80357410")
```

## 4. Interpretation and QC
- **Reproducibility:** Record the exact tool versions, transcript set, and genome build used.
- **HGVS:** Do not hand-derive HGVS strings from VCF POS. Rely on the engine's output, but verify 3'-shifting for indels.
- **Concordance:** If results differ between tools, check the transcript model and engine version. Discordance is common in indels and splice regions.

## 5. Output
Ensure the final VCF contains:
- Consequence type (SO term)
- HGVS cDNA and Protein changes
- Population frequencies (gnomAD, 1000G)
- Clinical significance (ClinVar, ACMG criteria if available)
- Pathogenicity scores (CADD, REVEL, SIFT, PolyPhen)