Zygosity

Zygosity is the relationship between the two copies of a locus in a diploid genome: homozygous when both copies carry the same allele, heterozygous when they differ. In your own data it is not an observation but an inference, made per site, from the reads that aligned there.

How it works

You have two copies of most of your genome, one from each parent. At a given position, a short-read caller collects the bases spanning that position and asks which of three models best explains them: 0/0 (both reference), 0/1 (one reference, one alternate), 1/1 (both alternate). With 30x coverage and a true heterozygote, you expect roughly 15 reads supporting each allele, binomially distributed. The caller converts that into genotype likelihoods (the PL field) and emits the most likely call.

Zygosity matters because inheritance modes are defined in terms of it. Recessive conditions require both copies to be disrupted, which can happen as a homozygous variant (the same allele from both parents, common in consanguineous pedigrees or on shared haplotypes) or as a compound heterozygote, two different damaging variants on opposite chromosomes. Lipoprotein lipase deficiency is a clean example of this structure: the severe phenotype requires biallelic loss, while single heterozygous frameshifts sit in a much milder and more variable range ¹. Dominant conditions need only one altered copy, and dosage-sensitive genes can show intermediate effects from a single hit. A promoter variant in ERAP1 reducing expression from one allele illustrates how a heterozygous regulatory change can still produce measurable consequences ².

“Is homozygous good or bad” has no general answer. Most of your homozygous sites are homozygous reference, which is unremarkable. Across populations, genome-wide heterozygosity does correlate weakly with fitness measures, but the meta-analytic effect is small and depends on identity disequilibrium in the sampled population ³. Those correlations often reflect inbreeding history across the whole genome rather than the specific markers typed ⁴. None of this transfers cleanly to a single person’s health.

In your own data

Zygosity lives in the GT field of your VCF. Read it alongside two neighbors:

• AD — allelic depth, reference and alternate read counts.
• DP — total depth at the site.

A 0/1 call backed by AD=28,2 is not a heterozygote. That is noise, an alignment artifact, or somatic mosaicism. A 1/1 call with DP=4 is a coin flip. We filter with something like:

bcftools view -i 'FMT/DP>=10 & FMT/GQ>=20' sample.vcf.gz \
  | bcftools query -f '%CHROM\t%POS\t[%GT\t%AD\t%GQ]\n'

Then compute allele balance for every het call and plot the distribution. A healthy WGS sample gives a tight peak near 0.5. A long tail toward 0.1–0.3 means you are looking at paralogous alignments (segmental duplications, pseudogenes like PMS2CL or SMN1/SMN2) where reads from two loci pile onto one reference position and manufacture false heterozygotes. Mark those regions and stop trusting calls inside them.

Common mistakes:

• Treating two heterozygous variants in the same gene as a compound heterozygote without phase. They may be on the same chromosome, which leaves one intact copy. Short reads phase variants within a few hundred bases at best. Long reads or parental genotypes resolve the rest.
• Ignoring that intermediate allele fractions can be real biology. Clonal hematopoiesis and tumor subclones produce variants at 5–30% VAF that no diploid genotyper will call correctly; single-cell genotyping in AML shows how much clonal structure a bulk average hides ⁵.
• Assuming heterozygous means half expression. Allelic expression is stochastic and regulatory heterozygotes do not simply average their two alleles ⁶. Check RNA-seq allele-specific counts at the same positions rather than assuming.
• Calling zygosity inside a deletion. A hemizygous site with one copy deleted looks homozygous to a diploid caller. Cross-check your CNV calls before interpreting any run of homozygosity.

Targeted assays exist when you need a single site called cleanly and cheaply, from classic allele-specific PCR designs ⁷ to newer extraction-free probe chemistries that discriminate zygosity directly ⁸. For confirming a clinically relevant genotype from WGS, orthogonal validation is still the standard, and interpretation of any result that could affect your care belongs with a clinical geneticist or genetic counselor.

Limitations

Zygosity is defined per reference position, which breaks down where the reference is a poor model of you: HLA, the immunoglobulin loci, highly repetitive regions, and structural variants. Loss of heterozygosity across a chromosome arm means something different in a tumor than in germline. And the diploid assumption itself is contingent. Hybrid genomes retain heterozygosity in ways that shape which alleles can be selected ⁹, and organisms like Candida albicans carry heterozygosity as a signature of hybrid ancestry rather than a per-site property ¹⁰.

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.