SNV and Indels

This section describes the SNV and indel calling steps in the Poppy pipeline. Starting from the sorted, duplicate‑marked BAM produced by the alignment module, two variant callers are run in parallel — GATK Mutect2 and VarDict. Their call sets are combined into a single ensemble VCF with bcbio‑variation‑recall, which is then decomposed, normalised, annotated (VEP, artifact panel, background panel), and filtered to produce the final somatic VCF.

A separate germline‑filtered branch of the Mutect2 calls is also created and used as input by the CNV module (PureCN / CNVkit).

Rules are provided by the Hydra‑Genetics snv_indels module (v0.6.0), the annotation module (v1.0.0), and the filtering module (v0.3.0).

SNV and Indel Workflow

Input Files

Input	Source
`alignment/samtools_merge_bam/{sample}_{type}.bam`	Alignment module
`alignment/samtools_merge_bam/{sample}_{type}.bam.bai`	Alignment module

Workflow Steps

1. GATK Mutect2 — Variant Calling

Somatic variant calling is performed per chromosome (using the same split strategy as the alignment module) with GATK Mutect2 in tumour‑only mode.

Item	Value
Container	`hydragenetics/gatk4:4.1.9.0`
Input	`alignment/samtools_merge_bam/{sample}_{type}.bam`
Output	`snv_indels/gatk_mutect2/{sample}_{type}.merged.vcf.gz`

2. VarDict — Variant Calling

VarDict is run in parallel as a second variant caller. It is configured with --nosv (structural variant calling disabled) and a minimum base quality of 1.

Item	Value
Container	`hydragenetics/vardict:1.8.3`
Input	`alignment/samtools_merge_bam/{sample}_{type}.bam`
Output	`snv_indels/vardict/{sample}_{type}.merged.vcf.gz`

3. bcbio‑variation‑recall Ensemble — Caller Consensus

The Mutect2 and VarDict VCFs are merged into a single ensemble VCF. A variant is included in the ensemble if it is called by at least one of the two callers.

Item	Value
Container	`hydragenetics/bcbio-vc:0.2.6`
Input	Per‑caller `{sample}_{type}.merged.vcf.gz` files
Output	`snv_indels/bcbio_variation_recall_ensemble/{sample}_{type}.ensembled.vcf.gz`

4. vt Decompose — Multi‑Allelic Splitting

Complex and multi‑allelic variants in the ensemble VCF are decomposed into separate records using vt decompose.

Item	Value
Container	`hydragenetics/vt:2015.11.10`
Output	`snv_indels/bcbio_variation_recall_ensemble/{sample}_{type}.ensembled.decomposed.vcf.gz`

5. vt Normalize — Variant Normalisation

Left‑alignment and trimming of indels against the reference genome using vt normalize.

Item	Value
Container	`hydragenetics/vt:2015.11.10`
Output	`snv_indels/bcbio_variation_recall_ensemble/{sample}_{type}.ensembled.decomposed.normalized.vcf.gz`

6. VEP — Variant Effect Prediction

All variants are annotated with functional consequences, population allele frequencies, and clinical significance using Ensembl VEP.

Item	Value
Container	`hydragenetics/vep:111.0`
Output	`snv_indels/bcbio_variation_recall_ensemble/{sample}_{type}.ensembled.vep_annotated.vcf.gz`

Key VEP options configured in Poppy:

--pick with ordering: MANE Select → MANE Plus Clinical → Canonical → Biotype → Rank → APPRIS → TSL → CCDS → Length
--check_existing — flags known variants
--everything — includes population frequencies (gnomAD, 1000G), SIFT, PolyPhen, etc.

7. Artifact Annotation — Panel‑of‑Normals Flagging

Positions that appear recurrently in normal samples (from a site‑specific panel of normals) are annotated with artifact counts and standard‑deviation scores. These fields are used by the soft‑filter step downstream.

Item	Value
Input	Annotated VCF + `reference.artifacts` panel
Output	`…vep_annotated.artifact_annotated.vcf.gz`

8. Background Annotation — Sequencing Background Flagging

Per‑position background noise levels (computed from the panel of normals) are added to the INFO field. Variants where the observed signal is close to the noise floor are flagged.

Item	Value
Input	Artifact‑annotated VCF + `reference.background` panel
Output	`…artifact_annotated.background_annotated.vcf.gz`

9. Hard Filtering — Somatic

Variants failing basic quality thresholds are removed (hard filter). The criteria are defined in config_hard_filter_somatic.yaml:

Filter	Criterion	Description
VAF	`AF < 0.01`	Variant allele frequency below 1 %
Depth	`DP < 100`	Total depth below 100 reads
AD	`AD[alt] < 5`	Fewer than 5 reads supporting the alt

Item	Value
Output	`…background_annotated.filter.somatic_hard.vcf.gz`

10. Soft Filtering — Somatic

Variants that pass the hard filter are evaluated against additional criteria and flagged (not removed) in the FILTER column. The criteria are defined in config_soft_filter_somatic.yaml:

Flag	Criterion
`Intron`	Intronic variant (unless also splice‑affecting)
`Consequence`	Non‑coding consequence (intergenic, UTR, regulatory, etc.)
`PopAF_0.02`	MAX_AF > 2 % in gnomAD / 1000G / ESP
`Biotype`	Not annotated as protein‑coding
`Artifact_gt_3`	Seen in ≥ 4 normals and < 5 SD from normal median AF
`Background_lt_4`	Position noise < 4 SD from median background

Item	Value
Output	`…filter.somatic_hard.filter.somatic.vcf.gz` (final somatic VCF)

Germline Branch (for CNV Module)

A separate filtering path is applied to the Mutect2‑only calls (not the ensemble) to produce a germline SNP VCF used by PureCN and CNVkit:

vt normalize → left‑align and trim
samtools sort → coordinate sort
VEP → annotate
Hard filter (germline) — keep common germline SNPs (gnomADe_AF ≥ 0.001, AD ≥ 50, SNV only)
bcftools annotate — add gnomAD annotations from small_exac_common_3.hg19.vcf.gz

Item	Value
Output	`snv_indels/gatk_mutect2/{sample}_{type}.normalized.sorted.vep_annotated.filter.germline.bcftools_annotated.vcf.gz`

Key Output Files

Output File	Description
`snv_indels/bcbio_variation_recall_ensemble/{sample}_{type}.ensembled.…filter.somatic.vcf.gz`	Final annotated and soft‑filtered somatic VCF
`snv_indels/gatk_mutect2/{sample}_{type}.merged.vcf.gz`	Mutect2 caller‑specific VCF
`snv_indels/vardict/{sample}_{type}.merged.vcf.gz`	VarDict caller‑specific VCF
`snv_indels/gatk_mutect2/{sample}_{type}.…filter.germline.bcftools_annotated.vcf.gz`	Germline SNP VCF (used by CNV module)

Downstream Consumers

The final somatic VCF and per‑caller VCFs are copied into the results/vcf/ output directory:

vcf/{sample}_{type}.filter.somatic.vcf.gz — ensemble somatic VCF
vcf/{sample}_{type}.gatk_mutect2.vcf.gz — Mutect2 raw calls
vcf/{sample}_{type}.vardict.vcf.gz — VarDict raw calls

The germline SNP VCF feeds into:

CNV module — PureCN (purity / ploidy estimation) and CNVkit (B‑allele frequency for CNV calling)

Configuration

The relevant sections in config.yaml:

bcbio_variation_recall_ensemble:
  container: "docker://hydragenetics/bcbio-vc:0.2.6"
  callers:
    - gatk_mutect2
    - vardict

gatk_mutect2:
  container: "docker://hydragenetics/gatk4:4.1.9.0"

vardict:
  container: "docker://hydragenetics/vardict:1.8.3"
  extra: " -Q 1 --nosv "
  bed_columns: "-c 1 -S 2 -E 3"

vep:
  container: "docker://hydragenetics/vep:111.0"
  mode: "--offline --cache --merged "
  extra: " --assembly GRCh38 --check_existing --pick --variant_class --everything …"

vt_decompose:
  container: "docker://hydragenetics/vt:2015.11.10"

vt_normalize:
  container: "docker://hydragenetics/vt:2015.11.10"

filter_vcf:
  germline: "config/config_hard_filter_germline.yaml"
  somatic: "config/config_soft_filter_somatic.yaml"
  somatic_hard: "config/config_hard_filter_somatic.yaml"

See the full config.yaml for all available settings.