variant-annotation

Variant Annotation

Annotation and clinical interpretation of DNA variants in cancer. Covers VCF normalization and filtering, functional annotation with VEP, germline classification under ACMG/AMP, somatic classification under the AMP/ASCO/CAP tiers and the ClinGen/CGC/VICC oncogenicity standard, tumor mutational burden, microsatellite instability, and neoantigen prediction.

When to Use This Skill

Activate when the user requests:

• VCF filtering, normalization, or merging
• Functional annotation (VEP, SnpEff, ANNOVAR)
• Germline variant classification (ACMG/AMP, pathogenic vs benign)
• Somatic variant classification (tiers, oncogenicity, actionability)
• Driver vs passenger discrimination
• OncoKB or CIViC evidence lookup
• Tumor mutational burden calculation
• Microsatellite instability estimation
• Neoantigen prediction or HLA binding
• Interpreting a variant of uncertain significance

Inputs

Data Type	Format	Source
Variant calls	VCF 4.2+, bgzipped and indexed	GATK, Mutect2, DeepVariant, Strelka2
Reference genome	FASTA with .fai	GRCh38 (or GRCh37 for legacy cohorts)
Somatic MAF	Mutation Annotation Format	TCGA GDC, cBioPortal
Panel definition	BED	Assay manufacturer, for TMB denominators
HLA type	4-digit alleles	OptiType, HLA-HD, arcasHLA

---

Environment

Versions verified 2026-08.

bcftools            1.24        VCF manipulation, normalization
Ensembl VEP         116.1       functional annotation, the current default
SnpEff              5.4c        alternative annotator
ANNOVAR             current     still distributed, registration required

Bioconductor
  VariantAnnotation 1.58.0      VCF parsing and coding consequences in R
  maftools          2.28.0      somatic MAF summarization, TMB, signatures

Neoantigens
  pVACtools         7.1.2       neoantigen pipeline (griffithlab)
  NetMHCpan         4.1+        MHC class I binding; academic licence

Microsatellite instability, and which tool is still alive:

  msisensor       ARCHIVED. ding-lab/msisensor was archived, last push 2021.
                  Do not start new work on it.
  msisensor-pro   v1.3.0 (2024-09). Current choice, tumor-only or paired.
  msisensor2      last push 2024-04. Tumor-only, no matched normal.
  MANTIS          paired tumor/normal, still used in clinical settings.

All four are read-based. MSI can also be inferred from a panel using
MSIsensor-pro's baseline mode, but the baseline must be built on the same
assay or the calls are not comparable.