Interpreting High and Low Heterozygosity

Why an Extreme Heterozygosity Value Needs Interpretation

A sample with unusually high or low heterozygosity is a diagnostic signal, not a diagnosis. A high value can arise when DNA from more than one individual contributes to a sample, but it can also reflect genotype-calling errors or a genuine difference in genetic background. Low heterozygosity can be consistent with inbreeding, yet it may also result from poor calls, selective missingness, or comparing a sample with an unsuitable reference group.

The interpretation depends on what was measured and which samples and variants form the comparison. A sample-level heterozygosity rate is usually calculated over a selected set of autosomal variants with calls passing specified filters. The selected variants, the call process, and the ancestry of the samples all affect the expected rate. Therefore, an extreme value should prompt targeted checks rather than automatic removal.

What the Heterozygosity Rate Does—and Does Not—Represent

At a biallelic locus with allele frequencies \(p_j\) and \(q_j=1-p_j\), Hardy–Weinberg equilibrium gives the true genotype probabilities \(p_j^2\), \(2p_jq_j\), and \(q_j^2\). Thus the probability that a true genotype is heterozygous is \(2p_jq_j\). This statement concerns true genotypes. It does not, by itself, specify the distribution of observed calls.

For observed data, let \(C_{ij}=1\) indicate that a genotype is called, and let \(O_{ij}\) be the call when it is called. A call can be wrong, and the probability of a call can depend on the true genotype. The expected heterozygosity among called genotypes therefore depends on the calling and missingness process:

Only with accurate calls and non-selective calling does the observed rate represent the true-genotype rate in the expected way. For example, if true genotypes are in Hardy–Weinberg equilibrium but the caller misses every true heterozygote, the observed heterozygosity among called genotypes is zero. Hardy–Weinberg equilibrium alone cannot prevent that result.

The expected true heterozygosity also depends on allele frequencies. At a locus, \(2p_jq_j\) is largest when the alleles are equally frequent and smaller when one is rare. A sample-level expectation averages across the selected loci, so its value depends on which variants were included and on the allele frequencies in the relevant population. This is why comparisons should use the same variant set and an appropriate reference group.

Interpreting High Heterozygosity

Contamination is one technical explanation for an unusually high observed rate. If DNA from another person is mixed with the sample, sites where the two individuals carry different homozygous genotypes can show evidence of both alleles. Depending on the assay and calling procedure, some of these sites may be called heterozygous in the mixed sample. The amount of the increase is not determined by contamination fraction alone: it also depends on the contaminant’s genotypes, the recipient’s genotypes, the platform, and the caller’s thresholds.

Poor calls can also produce an excess of apparent heterozygotes. For example, ambiguous intensity measurements or allele-balance patterns may lead a caller to assign a heterozygous genotype where a homozygous genotype is more plausible. But poor quality does not always raise heterozygosity: errors may instead cause missing calls or incorrect homozygous calls. Inspect call rate and genotype-level evidence rather than assuming one direction of bias.

Ancestry can also matter. The frequency of each allele, and hence expected heterozygosity at each locus, can differ across ancestry groups. A sample compared with a group whose expected heterozygosity is lower may appear unusually high even when its calls are sound. A biological difference is not automatically a technical failure, and an ancestry difference should not be treated as evidence of contamination.

Interpreting Low Heterozygosity

Inbreeding can reduce true heterozygosity relative to the expectation based on allele frequencies. Under the standard inbreeding model at a biallelic locus, with inbreeding coefficient \(F\), the true heterozygote probability is \(2p_jq_j(1-F)\). This describes a population model; it does not mean that an individual sample’s heterozygosity rate directly measures its family history without uncertainty or appropriate frequency estimates.

A low observed rate may instead arise from failed or selectively missing heterozygote calls, low-quality genotypes, or a mismatch between the sample’s ancestry and the frequencies used to calculate expected heterozygosity. A population with different allele frequencies can have a different expected rate even if its members are not inbred. In particular, comparisons based on an inappropriate pooled or subgroup-specific baseline can make an ordinary sample appear extreme.

A useful interpretation asks whether the low rate persists when the sample is compared with samples of similar ancestry using the same well-performing variants. Then check whether call rate, genotype quality, and missingness patterns support reliable measurement. If these checks are satisfactory, inbreeding may be a reasonable biological explanation to investigate, but heterozygosity alone does not prove it.

Why Ancestry-Matched Comparisons Matter

For an illustrative biallelic locus, a group with allele frequency \(p=0.10\) has expected true heterozygosity \(2(0.10)(0.90)=0.1800\) under Hardy–Weinberg equilibrium. A group with \(p=0.40\) has expected true heterozygosity \(2(0.40)(0.60)=0.4800\). Both calculations assume the stated frequencies and equilibrium; they do not imply that every individual has those rates.

Across many loci, differing frequencies can shift the average expected heterozygosity. A sample’s rate may therefore look low relative to one group and ordinary relative to another. This is a comparison problem, not evidence that ancestry itself is a quality defect. Conversely, pooling distinct groups can complicate interpretation because a single pooled frequency baseline may not represent either group well. The practical response is to use a defensible comparison set and to report how that set was formed.

A Practical Interpretation Sequence

Use heterozygosity as one component of sample-level quality control. First verify that the rate was calculated on an appropriate, consistently filtered set of variants and that the comparison set is relevant. Then evaluate whether the observed calls are sufficiently reliable to support a biological interpretation.

Deriving the Inbreeding-Model Relationship

The relationship used in the low-heterozygosity example follows from the standard inbreeding model. At a biallelic locus, let the allele frequencies be \(p_j\) and \(q_j=1-p_j\). Under the model, the genotype probabilities are \(p_j^2+F p_jq_j\) for one homozygote, \(2p_jq_j(1-F)\) for the heterozygote, and \(q_j^2+F p_jq_j\) for the other homozygote. The three probabilities sum to one, and the heterozygote probability is therefore reduced from its Hardy–Weinberg value by the factor \(1-F\).

Rearranging gives \(F=1-H_{\mathrm{obs}}/H_{\mathrm{exp}}\) when the observed and expected quantities are valid estimates under the same model and comparison. In a sample-level calculation, this ratio is a summary across loci; departures caused by call errors, frequency mismatch, or variant selection can also change it. The algebra is exact for the model, while the interpretation depends on its assumptions.

Diagnostics, Alternatives, and Limitations

Call rate is useful alongside heterozygosity, but it answers a different question: what fraction of eligible genotypes were called? A good call rate does not guarantee accurate genotypes, and a low rate does not by itself explain whether heterozygosity is high or low. Genotype-level quality measures can help distinguish these possibilities. Hardy–Weinberg diagnostics can flag locus-level patterns, but they are not a stand-alone explanation for an individual sample’s heterozygosity; population structure, ascertainment, and technical errors can affect those patterns as well.

A common mistake is to reason, “This sample has high heterozygosity, so it is contaminated,” or “This sample has low heterozygosity, so it is inbred.” The correction is to treat each statement as a hypothesis and seek independent evidence. High heterozygosity with unusual allele balance and other quality concerns supports a technical explanation more than high heterozygosity alone. Low heterozygosity with sound calls and an appropriate ancestry-matched baseline is more consistent with a biological difference, but still does not prove its cause.

Heterozygosity is also sensitive to variant selection, allele-frequency estimation, and missingness. Different filters can yield different rates, so values from different variant sets should not be compared as if they were directly interchangeable. Finally, no single diagnostic perfectly separates contamination, calling error, inbreeding, and ancestry differences. The strongest interpretation combines the heterozygosity result with appropriate comparison samples and independent technical evidence.