Why an Array Does Not Represent Every Variant Equally
A genotyping array measures the alleles at a set of selected genomic positions. The earlier material on how genotyping arrays measure alleles and probe design errors explains how a designed marker can be interrogated and how a probe can fail. A separate issue arises before measurement: which variants were selected for the array in the first place? If selection depends on which variants were known, common, or technically suitable in a particular discovery panel, the array’s contents will not be a neutral sample of all genomic variation.
This selection process is called variant ascertainment. It matters because an array can measure its included markers accurately while still giving uneven information about populations or genomic regions. A variant missing from the design cannot be directly measured by that array, no matter how well the included probes perform.
Discovery Panels and the Chance of Finding a Variant
Array designs often begin with a discovery panel: a group of individuals in whom variants are identified and their frequencies estimated. Let \(q\) be a variant’s allele frequency in the population represented by that panel, and let \(m\) be the number of sampled chromosome copies. Under random sampling with independent allele copies, the probability that a particular allele copy is not the variant is \(1-q\). The probability that the variant is absent from all \(m\) copies is therefore \((1-q)^m\). Thus, the probability of observing the variant at least once is:
This calculation assumes the sampled chromosome copies behave as independent draws with the same allele frequency \(q\). For a panel of \(N\) diploid individuals, \(m=2N\) chromosome copies under this approximation. The probability depends strongly on \(q\): a common variant is likely to appear, while a rare variant may be missed even in a sizeable panel. If a design requires at least \(k\) copies of the variant rather than one, the probability of satisfying that rule is the binomial tail \(P(X\geq k)\), where \(X\sim \mathrm{Binomial}(m,q)\). A more restrictive threshold makes discovery of rare variants still less likely.
A further selection step may favor variants above a minor allele frequency threshold in the discovery panel. Such a threshold can be useful for a design aimed at common variation, but it defines the array’s scope: variants that are rare or absent in that panel are less likely to be included. Frequencies can differ between populations, so a variant common in the discovery panel may be rare elsewhere, and a variant common elsewhere may not have been discovered at all.
Worked Example: A Common-Variant Inclusion Rule
Suppose a hypothetical discovery panel contains 50 diploid individuals, so it contributes 100 chromosome copies. A design team includes known variants whose observed minor allele frequency in this panel is at least 5%. One candidate variant has 18 copies of its less frequent allele; another has 3. Apply the rule.
For the first candidate, the observed minor allele frequency is \(18/100=0.18\), or 18%. Since \(0.18\geq0.05\), it passes the frequency rule. For the second, it is \(3/100=0.03\), or 3%. Since \(0.03<0.05\), it does not pass. On this rule alone, the first candidate is eligible and the second is not.
Now suppose, hypothetically, that the first variant has frequency 0.16 in a second population and 0.01 in a third. Its inclusion does not mean it is common in both populations: the same array marker can be frequent in one and rare in another. Conversely, the exclusion of the second candidate does not establish that it is unimportant in the third population; the panel-based threshold simply did not select it.
Common-Variant and Rare-Variant Ascertainment
A common-variant array typically selects from variants already discovered and observed frequently enough in its discovery resources to meet the design’s criteria. Its markers may be useful for studying common variation in populations similar to those represented in discovery, but the resulting set is not a random sample of genomic variants. In particular, its frequency distribution is enriched for variants that were detectable and sufficiently common in those resources.
Rare-variant selection has a different challenge. A rare variant may not appear in a discovery panel simply because few chromosome copies were sampled. A design specifically targeting rare variants can use larger or more diverse discovery resources, or focus on previously identified variants of interest. But it remains a selected set: targeted rare-variant content does not make an array an exhaustive survey of all rare variation.
The following calculation isolates the discovery issue. It does not model additional filters, such as whether a discovered variant is chosen for the final design.
Worked Example: How Often Is a Rare Variant Missed?
Assume a hypothetical variant has frequency \(q=0.01\) and a discovery panel contributes 60 independently sampled chromosome copies. Under the sampling model above, the probability it is observed at least once is:
So the chance of observing it at least once is about 45.3%; the chance of missing it is \(0.99^{60}\approx0.5472\), or about 54.7%. This is true even though the variant exists in the population. For comparison, if \(q=0.05\), the probability of observing it at least once is \(1-0.95^{60}\approx0.9539\), or about 95.4%. The difference follows from the sampling probabilities, not from any difference in probe performance.
These probabilities concern detection at least once. If the design requires multiple observed copies, or excludes variants below a frequency threshold, the chance that a rare variant is ultimately included is lower still.
Uneven Representation Across Populations and Regions
Array content can be uneven in two related but distinct ways. Population representation concerns whether the selected variants are informative about variation in a population of interest. Regional representation concerns how densely selected markers cover different parts of the genome or a chosen catalog of variants. A design can have many markers overall and still represent a particular population or region poorly.
Population differences arise partly because variant frequencies and histories differ among populations, while discovery panels and selection rules are finite. A variant selected for being common in one panel may be rare or absent in another; variants common in the latter may not have entered the candidate list. This does not imply that populations are internally uniform or that a broad population label determines an individual’s genotype. It means that the design and discovery resources should be evaluated against the actual populations and questions of interest.
Regional differences can arise because known variants are not distributed uniformly across the genome, and because an array cannot include every candidate. Design choices, available variant catalogs, and technical constraints can leave some regions with many selected markers and others with few. The relevant denominator must be stated: a count of array markers per region is not the same as the fraction of known variants in that region represented by the array.
Worked Example: Comparing Representation Across Regions
Consider three hypothetical genomic regions, each 1 megabase long. A chosen catalog lists 20 eligible variants in each region. The array includes 12 catalog variants in Region A, 8 in Region B, and 2 in Region C. Calculate the fraction of catalog variants represented in each region.
For Region A, \(12/20=0.60\), or 60%. For Region B, \(8/20=0.40\), or 40%. For Region C, \(2/20=0.10\), or 10%. Because the regions have equal length, the included-marker densities are also 12, 8, and 2 markers per megabase, respectively.
The array therefore represents this stated catalog much more densely in Region A than in Region C. The conclusion is limited to the catalog and eligibility definition used here; it does not establish that Region C contains less biological variation. If another population has a different set of variants, these percentages may not describe its representation.
Assessing an Array’s Ascertainment
A useful assessment begins by making the target explicit. Ask which populations, variant-frequency range, regions, and biological questions the design is intended to serve. Then compare the design’s variant list with relevant reference data or a clearly defined catalog. Summary statistics should preserve the denominator and grouping used in the comparison; otherwise, apparently precise coverage percentages can conceal what is and is not represented.
Record which panels or resources supplied candidate variants, and what frequency or other selection rules were applied.
Specify the populations, variant classes, regions, or catalog against which representation will be judged.
For example, calculate the fraction of eligible catalog variants included separately by region or population, and report each denominator.
Check whether a variant is absent from the design before attributing its lack of a genotype result to a measurement problem. Included markers can also have probe-level problems, a distinct issue.
Potential warning signs include a strong difference in the array’s variant-frequency distribution from the target reference, low inclusion fractions for particular regions, or large differences in marker representation across target populations. These are diagnostics of possible design limitations, not automatic proof that every analysis using the array is invalid. The consequences depend on the scientific question and on which variants are required to answer it.
Design Choices, Alternatives, and Limitations
Different design strategies address different goals. A common-variant design can prioritize variants frequent in a specified discovery resource. A broader discovery strategy that includes multiple populations can reduce dependence on any single panel, though it cannot guarantee equal representation everywhere. A targeted design can add variants of specific interest, including rare variants, but its content is limited to those selected targets. These strategies change which variants are available; none makes a finite array a complete inventory of genomic variation.
| Design emphasis | Potential strength | Important limitation |
|---|---|---|
| Common variants from one discovery panel | Content is aligned with common variation observed in that panel | Variants uncommon or absent there can be missed; representation may differ elsewhere |
| Discovery across multiple populations | Can broaden the range of variants considered for inclusion | Finite resources and design capacity still require selection; inclusion is not automatically balanced |
| Targeted rare-variant content | Can include specified rare variants that common-variant rules would tend to omit | Does not capture rare variants that were not discovered or selected |
The central limitation is structural: an array measures only its chosen markers. Frequency differences, incomplete discovery, and regional design constraints can all affect what variation is observable. A good report therefore describes the array’s ascertainment basis and the target against which representation was evaluated, rather than treating “genome-wide” as a guarantee of uniform coverage.
Common Mistakes
- “The array has many markers, so it represents every population well.” A large total marker count does not establish that the markers are informative or evenly distributed for every target population.
- “A variant absent from the array is absent from the population.” Absence from the design means it was not directly assayed; it is not evidence that the variant does not exist.
- “Low representation in a region means the region has little variation.” The number of selected markers reflects design and ascertainment as well as the available catalog. Interpret coverage only against a stated denominator.
- “A rare-variant array captures rare variation generally.” It captures the rare variants selected for that design, not every rare variant in the genome.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- A discovery panel samples 40 diploid individuals. Under the independent chromosome-copy model, what is the probability of observing a variant with frequency 0.02 at least once?
- Why can a variant selected as common in a discovery panel be rare in a different population?
- An equal-length region contains 30 eligible catalog variants, of which an array includes 9. What fraction is represented, and what does that fraction refer to?
- Does a targeted rare-variant array measure all rare variants? Explain.
Solutions. (1) There are \(m=2(40)=80\) chromosome copies, so the probability is \(1-(1-0.02)^{80}=1-0.98^{80}\approx1-0.1986=0.8014\), about 80.1%. This is the probability of at least one observation under the stated sampling model, not a guarantee of inclusion on the final array. (2) Allele frequencies can differ among populations, and selection used the discovery panel’s frequency; being common there does not require being common elsewhere. (3) The fraction is \(9/30=0.30\), or 30%. It is the proportion of the specified eligible catalog variants in that region represented by the array, not a claim about all variation in the region. (4) No. The array measures only rare variants that were discovered and selected for its design; other rare variants may be absent from its content.