Population-Specific Frequency and Rarity Classification

Why Rarity Depends on the Population Being Described

A variant’s allele frequency is a proportion in a defined set of chromosomes, not an intrinsic label that is identical in every population. If a variant is uncommon in one ancestry group and more frequent in another, a pooled estimate depends on how many chromosomes from each group were included. That estimate may be useful for describing the sample, yet misleading for a person or target population whose ancestry composition differs from the sample.

This matters because rarity is often assigned using a frequency threshold. A threshold can classify a variant as rare in one group but not another, or a pooled frequency can conceal a higher frequency in a subgroup. Such a classification describes frequency under a stated population definition and threshold. It does not, by itself, establish whether a variant is pathogenic or clinically important.

Group-Specific and Pooled Frequencies

Definition: For a specified group \(g\), the allele frequency \(p_g\) is the proportion of observed allele copies in that group that are the allele of interest. In a fully genotyped sample of \(n_g\) diploid individuals at an autosomal locus, if \(a_g\) copies of the allele are observed, then \(\hat{p}_g=a_g/(2n_g)\). A pooled frequency combines allele copies across groups; it is not generally the unweighted average of the group-specific frequencies.

The denominator is the number of observed allele copies, not simply the number of people. For fully called diploid autosomal genotypes, each individual contributes two copies. If the number of callable chromosomes differs across individuals or groups, the appropriate denominator is the number of callable copies in each group. Other inheritance patterns or ploidy require the corresponding number of observed allele copies rather than automatically using \(2n_g\).

“Ancestry group” needs a clear operational meaning in any analysis. It may refer to a study-defined genetic ancestry group, a geographic sampling group, or another explicitly defined population category. These groupings are not fixed, sharply bounded biological types: people within a group vary, and group definitions may not capture an individual’s full ancestry. Frequency comparisons should therefore describe the groups actually sampled and defined, rather than imply that a label determines an individual’s genotype.

A rarity threshold is also a decision rule, not a property of the frequency calculation. For illustration, this tutorial will sometimes call a variant “rare” when its frequency is below 1%. A different analysis may use a different threshold, and the threshold should be stated along with the population to which it is applied.

Worked Example: Different Frequencies in Two Groups

Suppose a hypothetical study observes fully called genotypes at an autosomal locus in two ancestry groups, with 500 diploid participants in each. The allele of interest is observed twice in Group A and 24 times in Group B. Use a 1% threshold for the illustrative rarity classification.

For Group A there are \(2(500)=1{,}000\) observed allele copies, so:

\hat{p}_A=\frac{2}{1{,}000}=0.002=0.20\%.

For Group B there are also 1,000 observed copies:

\hat{p}_B=\frac{24}{1{,}000}=0.024=2.40\%.

Under the stated rule, the observed frequency is below 1% in Group A and above 1% in Group B. The conclusion is not that the allele has one universally applicable rarity status; it is that these sample estimates fall on opposite sides of the illustrative threshold in the two defined groups. The estimates also have sampling uncertainty, especially when allele counts are small.

How Population Mixture Determines a Pooled Frequency

Suppose a sample combines groups, and let \(w_g\) be the proportion of all observed allele copies contributed by group \(g\). The weights are nonnegative and sum to 1. The pooled frequency is the weighted average of the group-specific frequencies:

\hat{p}_{\mathrm{pool}}=\sum_g w_g\hat{p}_g, \qquad \sum_g w_g=1.

When every participant contributes the same number of callable allele copies, the weights are also the proportions of participants from each group. More generally, they are proportions of observed allele copies. This distinction matters if group sample sizes differ, genotypes have different missingness, or individuals contribute different numbers of copies.

The formula follows directly from adding allele counts. If group \(g\) contributes \(m_g\) observed allele copies and has frequency \(\hat{p}_g=a_g/m_g\), its allele count is \(a_g=m_g\hat{p}_g\). Across groups, the pooled frequency is:

\hat{p}_{\mathrm{pool}} =\frac{\sum_g a_g}{\sum_g m_g} =\frac{\sum_g m_g\hat{p}_g}{\sum_g m_g} =\sum_g \left(\frac{m_g}{\sum_h m_h}\right)\hat{p}_g.

Thus \(w_g=m_g/\sum_h m_h\). The pooled value describes the combined sample’s allele copies. It need not describe any one group, and it may not describe a different target population with different group proportions.

Worked Example: The Same Group Frequencies, Different Mixtures

Use the hypothetical frequencies from the previous example: \(\hat{p}_A=0.002\) and \(\hat{p}_B=0.024\). First suppose 90% of the observed allele copies come from Group A and 10% from Group B. The pooled frequency is:

\hat{p}_{\mathrm{pool}} =(0.90)(0.002)+(0.10)(0.024) =0.0018+0.0024 =0.0042=0.42\%.

This pooled estimate is below 1%, although Group B’s frequency is 2.40%. Now consider a sample in which 20% of the copies come from Group A and 80% from Group B:

\hat{p}_{\mathrm{pool}} =(0.20)(0.002)+(0.80)(0.024) =0.0004+0.0192 =0.0196=1.96\%.

The group-specific frequencies have not changed, but the pooled estimate has moved from 0.42% to 1.96% because the mixture changed. Under the illustrative 1% rule, the first pooled sample is classified as rare and the second is not. This shows why a pooled rarity label should be accompanied by the population composition used to calculate it.

When the Sample Does Not Represent the Target

A study’s pooled frequency is an appropriate estimate for the target only when the sampled chromosomes represent the target population for the purpose at hand. If a study has different group proportions from the target, its raw pooled frequency answers a different question. Where suitable group-specific estimates are available and the target group proportions are specified, a target-weighted frequency can be calculated. This calculation adjusts the mixture; it does not fix biased or inaccurate group-specific estimates.

A sample can be unrepresentative because some groups were recruited in greater or lesser proportions than they occur in the target, or because participation and data availability differ across groups. The direction of the resulting difference depends on how the group-specific frequencies and sample proportions compare with those in the target. Oversampling a group does not always increase the pooled frequency; it increases or decreases it according to that group’s frequency relative to the other groups.

Worked Example: Reweighting an Unrepresentative Sample

Suppose a hypothetical study estimates frequencies of 0.20% in Group A and 2.40% in Group B. Its sample is 95% Group A and 5% Group B, but the target population is 20% Group A and 80% Group B. Assume, for this illustration, that the group-specific estimates are suitable for both calculations and that each participant contributes the same number of callable copies. Use the 1% threshold.

The raw study mixture gives:

\hat{p}_{\mathrm{study}} =(0.95)(0.002)+(0.05)(0.024) =0.0019+0.0012 =0.0031=0.31\%.

The target-weighted estimate is:

\hat{p}_{\mathrm{target}} =(0.20)(0.002)+(0.80)(0.024) =0.0004+0.0192 =0.0196=1.96\%.

The study’s raw pooled estimate is below 1%, whereas the target-weighted estimate is above 1%. The difference is due to the group proportions used in the two calculations, not a change in either group’s estimated frequency. The target-weighted result remains conditional on the assumed group-specific estimates and target proportions being appropriate; it should not be presented as a directly observed frequency in the target population.

Interpreting Frequency Differences Carefully

When comparing estimates across groups, separate three questions: what frequency was estimated in each group, what population does a pooled estimate represent, and what threshold or interpretation rule is being applied? Keeping these questions distinct helps avoid treating a mixture-dependent summary as a universal property of a variant.

Quantity	What it describes	Interpretive caution
Group-specific frequency	Allele copies in a defined group or sample	Depends on the group definition and the observed sample
Pooled sample frequency	All allele copies combined in a particular sample	Depends on sample composition and callable-copy counts
Target-weighted frequency	A weighted combination using specified target proportions	Depends on the suitability of group estimates and target weights
Rarity classification	A frequency compared with a stated threshold	Can differ by population and threshold; does not establish pathogenicity

Frequency differences should be reported with their denominators and population definitions, not just percentages. A frequency based on a few observed allele copies may be imprecise; a small difference around a cutoff should not be treated as a certain distinction. The sampling uncertainty of rare allele counts is a separate issue from mixture weighting: weighting addresses which groups contribute to a summary, while uncertainty concerns how much the observed counts may vary across samples.

In practice, ancestry categories are imperfect summaries of complex histories and relatedness. Frequency estimates can vary within a category, and a person may not be well represented by a single group label. For individual interpretation, a group-level estimate is relevant evidence only to the extent that the reference group is appropriate for that individual and the estimate is sufficiently reliable. A rarity classification alone is not a causal explanation for a phenotype.

Common Mistake

Common mistake: “The variant is below 1% in the pooled sample, so it is rare in every ancestry group.” This does not follow. A pooled frequency can be below the cutoff while one group’s frequency is above it, as the 0.42% pooled estimate and 2.40% Group B estimate illustrate. The correction is to state the population represented by the pooled estimate, inspect group-specific estimates when relevant, and apply the stated threshold to the population that matches the question.

The reverse mistake is also possible: seeing a frequency above a cutoff in one group does not establish that the frequency is above the cutoff in every group. Neither pooled nor group-specific frequency should be detached from its denominator, group definition, sample composition, and uncertainty.

Key takeaway: Rarity classification is population-specific whenever allele frequencies differ among groups. A pooled frequency is a weighted average whose weights come from the observed allele-copy composition; it represents a target population only when its composition is appropriate or the estimate has been correctly reweighted.

Check Your Understanding

Use the definitions and calculations above to answer the following questions.

A fully called autosomal sample includes 400 diploid people from one group and has 8 copies of an allele. What is the estimated allele frequency, as a percentage?
Two groups have frequencies of 0.5% and 3%. What additional information is needed to calculate the pooled frequency for a combined sample?
Explain why a pooled estimate below a 1% threshold does not prove that the frequency is below 1% in every group.
A study’s sample is 80% Group A, but the target population is 30% Group A. Why might the raw pooled study frequency be a poor summary of the target population?
Does classifying a variant as rare in a specified group establish that it is pathogenic? Explain briefly.