Exploring a Personal Genome at 30x Depth

QC metrics: how to tell if a genome is good

Three numbers tell you most of what you need to know about a WGS callset:

• Ti/Tv ratio: transitions vs transversions. Expected value for WGS is ~2.0. A ratio well below 2.0 suggests contamination, alignment artefacts, or low-quality variant calls.
• Het/Hom ratio: heterozygous vs homozygous variant calls. Expected range for a single individual is 1.5 to 1.7. Values outside this range may indicate sample contamination (too high) or consanguinity (too low).
• Total variant count: a European genome typically carries 3.5 to 4.5 million SNPs relative to GRCh37. Counts far outside this range warrant investigation.

This genome: Ti/Tv = 2.03, Het/Hom = 1.63, total SNPs = 3,716,648. All within expected ranges.

Structural variants: the hidden layer

Structural variants (SVs) are genomic rearrangements of 50 bp or larger. They are harder to detect than SNPs and indels, require specialised callers, and are frequently missed by standard short-read pipelines. Yet SVs contribute disproportionately to rare disease.

SVs range in size from 50 bp to megabases. An estimated 20% of rare disease cases involve structural variants, but most clinical pipelines still focus primarily on SNPs and small indels.

From SNP chip to WGS: what changes clinically

Moving from a genotyping array to whole-genome sequencing does not just increase the variant count. It opens entire categories of analysis that are impossible with arrays:

• Full CYP2D6 haplotyping: gene deletions and duplications that change metaboliser status are invisible to SNP chips. WGS detects the structural rearrangements directly.
• HLA typing: the HLA region is too polymorphic and structurally complex for array-based genotyping. WGS reads through the full region.
• Novel variant discovery: arrays only test known positions. WGS finds variants that have never been catalogued before.
• Coding indels: frameshift insertions and deletions in protein-coding genes are a major source of loss-of-function. Arrays do not detect them.
• SVs disrupting gene function: large deletions or inversions that break genes are only visible with sequencing data.

The Corpasome: 13 years of open genomics

The Corpasome is one of the longest-running open personal genome projects:

• 2013: Manuel Corpas published his 23andMe genotype data under CC0 (public domain). One of the first fully open personal genomes. Used in dozens of studies and teaching exercises worldwide.
• 2026: The same individual, now sequenced at 30x depth by Dante Labs. Full WGS data released under CC0 on Zenodo.

Two datasets, same genome, 13 years apart. The 23andMe file has ~600,000 SNPs. The WGS file has 3.7 million SNPs, 912,000 indels, and 8,925 structural variants. Same person, radically different resolution.

Citations:
Corpas, M. (2013). Crowdsourcing the Corpasome. Source Code for Biology and Medicine, 8, 13. doi:10.1186/1751-0473-8-13
Corpas, M. (2026). 30x WGS of the Corpasome. Zenodo. doi:10.5281/zenodo.19297389

Setup (2 minutes)

Run the first two code cells. They clone the ClawBio repository and install dependencies (pysam, requests, pandas, matplotlib). You should see:

Load and explore the chr20 subset (5 minutes)

The notebook loads the chromosome 20 VCF subset. This is a manageable slice of the full genome, small enough to process quickly but large enough to demonstrate real WGS data. You will see:

• Total variant count on chr20 (SNPs + indels)
• Breakdown by variant type: SNPs vs insertions vs deletions
• Ti/Tv ratio for chr20 (should be close to 2.0)
• Het/Hom ratio for chr20 (should be close to 1.6)

Discussion point: Why is the Ti/Tv ratio a useful QC metric? (Transitions are biochemically more likely than transversions. A ratio near 2.0 indicates clean, unbiased variant calls.)

Explore structural variants (8 minutes)

Load the SV VCF and examine the types, sizes, and chromosomal distribution of structural variants across the genome.

• Count variants by type: DEL, DUP, INV, BND, INS
• Plot size distribution of deletions and duplications
• Identify the largest deletion and the largest duplication
• Check which genes overlap with large SVs

Discussion point: Why are deletions the most common SV type? (Deletions are easier to detect from short reads than insertions or complex rearrangements. This is partly biological and partly a detection bias.)

Extract pharmacogenomic variants from WGS (5 minutes)

Load the PGx loci subset extracted from the WGS data. Compare what the whole-genome sequence found at pharmacogenomic loci with what the original 23andMe chip reported.

• Load the 5 PGx variants from the WGS subset
• Cross-reference with the 23andMe findings from the previous workshop
• Identify any variants that WGS captured but the SNP chip missed
• Note differences in genotype calling between platforms

Annotate chr20 variants (5 minutes)

Run the variant-annotation skill on a selection of chr20 variants. The Ensembl VEP REST API has a limit of 200 variants per batch, so the notebook selects a clinically relevant subset.

• Select variants in coding regions on chr20
• Submit to VEP for functional annotation
• Retrieve ClinVar significance and gnomAD frequencies
• Review the prioritised output: Tier 1 through Tier 4

Exercises (15 minutes, independent work)

Three exercises for students: