Why Rare Alleles Need Explicit Uncertainty
A low-frequency allele can be easy to overlook in a virtual population: most individuals will not carry it, and the homozygous genotype may be exceptionally uncommon. Yet an allele’s frequency and its functional effect are separate pieces of information. A rare allele may have normal, decreased, no, or not-yet-characterized function. Treating “not characterized” as “normal” silently converts missing evidence into a biological claim.
The practical aim is not to force a definitive assignment when the evidence does not support one. Instead, preserve the uncertain annotation and ask how alternative, defensible functional assignments would affect the predicted phenotype proportions. This separates what is known about allele frequency from what remains uncertain about function.
Allele frequencies should refer to the population relevant to the virtual cohort. The earlier material on sourcing allele frequencies and interpreting population differences is important here: a rare allele in one population may be less rare in another, and a frequency estimate can itself be uncertain. This tutorial focuses on how to represent the functional uncertainty conditional on the chosen frequency.
From Allele Frequency to Genotype Proportions
For a transparent first calculation, consider a two-allele locus with reference allele \(*1\) at frequency \(1-q\) and rare allele \(*x\) at frequency \(q\). Under Hardy–Weinberg equilibrium, the two alleles in a genotype are treated as independent draws from the population gene pool. This approximation is appropriate when the population is adequately described by random mating at the locus and there is no substantial departure caused by factors such as population mixture, strong selection, or nonrandom mating.
These are genotype proportions, not functional classifications. To obtain phenotype proportions, each diplotype must be mapped to a phenotype under a stated functional assignment and a stated genotype-to-phenotype rule. The earlier material on deriving genotype frequencies and calculating pharmacogenetic phenotype frequencies supplies that framework; the key issue here is to repeat the calculation under alternative assignments when function is uncertain.
The rarity of \(*x\) does not mean its homozygotes can always be ignored. Their expected proportion is \(q^2\), which may be tiny, but a sufficiently large cohort can still include them. Whether they affect the result depends on the population size and on the phenotype rule.
Worked Example: A Rare Allele With an Unknown Effect
Suppose a hypothetical population has \(*x\) frequency \(q=0.006\). Assume Hardy–Weinberg equilibrium. For illustration, suppose \(*1/*1\) is normal function, one copy of a decreased-function allele produces an intermediate phenotype, and two copies produce a poor-function phenotype. Compare the prediction if \(*x\) is assigned decreased function with the prediction if it is assigned normal function.
First calculate genotype proportions:
If \(*x\) is assigned decreased function, the stated rule gives 98.8036% normal, 1.1928% intermediate, and 0.0036% poor function. If \(*x\) is assigned normal function, all three genotypes are classified as normal under this example’s rule, giving 100% normal function.
The difference between the scenarios is not evidence that one assignment is correct. It shows what the assignment changes: in this hypothetical population, the decreased-function scenario places 1.1964% of individuals in a non-normal phenotype category. In a cohort of 10,000, the corresponding expected counts are about 9,880 normal, 119 intermediate, and 0.36 poor under the decreased-function assignment. These are expected counts, not a guarantee that a finite simulated cohort will contain exactly those numbers.
Quantifying the Effect of Alternative Assignments
A useful sensitivity analysis holds the allele frequency and genotype assumptions fixed while changing only the functional assignment. That comparison identifies the contribution of the uncertain annotation, rather than mixing it with changes in population composition or in the phenotype definition.
The following short derivation explains a common result. If carrying \(*x\) is sufficient to place an individual in an affected category, then the affected proportion is the sum of the heterozygote and homozygote proportions. Under Hardy–Weinberg equilibrium:
If \(*x\) is instead assigned normal function, the affected proportion attributable to this allele is zero under that same illustrative rule. Thus the difference between the assignments is \(2q-q^2\), not \(q\): the carrier proportion includes both heterozygotes and homozygotes. This result depends on the dominant carrier rule in the example; a different genotype-to-phenotype rule gives a different comparison.
Worked Example: Comparing Decreased and No-Function Assignments
Consider another hypothetical single-locus population with \(q=0.010\), under Hardy–Weinberg equilibrium. To make the consequence of function explicit, use a simplified normalized activity score: each normal-function allele contributes 1, each decreased-function allele contributes 0.5, and each no-function allele contributes 0. Define normal phenotype as a total score at least 1.5, intermediate as a score from 0.5 up to but not including 1.5, and poor as a score below 0.5. These thresholds are illustrative and are not universal clinical cutoffs.
The genotype proportions are:
If \(*x\) has decreased function, the scores for \(*1/*1\), \(*1/*x\), and \(*x/*x\) are 2, 1.5, and 1, respectively. Therefore, 99.99% are normal and 0.01% are intermediate; none are poor under the stated cutoffs.
If \(*x\) has no function, the three scores are 2, 1, and 0. The phenotype proportions are then 98.01% normal, 1.98% intermediate, and 0.01% poor. The no-function assignment therefore lowers the normal proportion by 1.98 percentage points relative to the decreased-function assignment, under this particular scoring rule. The comparison is a consequence of both the functional assignment and the chosen phenotype cutoffs.
Worked Example: Bounding the Impact of an Uncertain Allele
Suppose a hypothetical population has two mutually exclusive non-reference alleles: a known no-function allele at frequency 0.002 and an allele of uncertain function at frequency 0.003. Assume Hardy–Weinberg equilibrium and treat the combined non-reference frequency as \(q=0.005\). For this example only, define the affected category as carrying at least one allele assigned no function.
If the uncertain allele is assigned normal function, only the known no-function allele contributes to the affected category. Its carrier proportion is:
or 0.3996%. If the uncertain allele is also assigned no function, any carrier of either allele is affected. The combined carrier proportion is:
or 0.9975%. The difference is \(0.009975-0.003996=0.005979\), or 0.5979 percentage points. This range is conditional on the assumed frequencies, Hardy–Weinberg equilibrium, and the carrier-based affected definition; it is not a confidence interval for the true population proportion.
Choosing and Checking an Uncertainty Representation
A single functional assignment is simplest, but it can conceal how much the output depends on uncertain evidence. When more than one assignment is plausible, report scenario-specific phenotype proportions, including the assumptions that define each scenario. If evidence supports a range of plausible functional classes, the resulting phenotype proportions can be presented as a sensitivity range rather than as one overly precise estimate.
A probabilistic functional assignment is another option when there is a defensible basis for assigning probabilities to functional classes. Such probabilities describe uncertainty about the allele’s function; they are not the allele frequency and should not be substituted for it. Without evidence to support those probabilities, scenario analysis is more transparent than inventing a precise probability.
| Approach | What it conveys | Main caution |
|---|---|---|
| Single assignment | Phenotype proportions conditional on one chosen functional class | Can imply greater certainty than the evidence supports |
| Alternative scenarios | How outputs change under distinct plausible assignments | Scenarios must use consistent frequency and phenotype assumptions |
| Probabilistic assignment | Uncertainty averaged across functional classes using specified probabilities | Requires defensible probabilities and clear reporting of their source |
Several checks help identify an inconsistent analysis. Confirm that allele frequencies are from the intended population and sum to no more than 1. Check that the genotype proportions sum to 1, and verify that each diplotype is assigned once—and only once—to a phenotype in each scenario. Compare the phenotype proportions with the underlying carrier proportion: a predicted affected proportion larger than the carrier proportion is impossible if only carriers of the allele can be affected.
Also inspect whether rare homozygotes have been omitted merely because their expected frequency is small. They may be negligible for a small cohort, but that should follow from the cohort size and the question being asked. Finally, distinguish a frequency estimate with sampling uncertainty from uncertainty about function. They are different uncertainties and may need separate sensitivity analyses.
Common Mistakes and Limitations
- Calling an uncharacterized allele normal. Lack of functional evidence does not establish normal function. Retain an uncertain annotation and compare justified alternatives.
- Using the allele frequency as the carrier proportion. Under Hardy–Weinberg equilibrium, the carrier proportion is \(2q-q^2\), not \(q\), when one or more copies suffice.
- Reporting scenario differences as biological certainty. A sensitivity result shows dependence on assumptions; it does not resolve which functional assignment is correct.
- Ignoring the phenotype rule. The same functional assignment can yield different phenotype proportions under different diplotype rules or cutoffs. State the rule before calculating.
These calculations are limited by the assumptions used to obtain genotype proportions and by the quality of the allele-frequency and functional evidence. Hardy–Weinberg proportions may not describe a population with substantial structure or nonrandom mating. Rare-allele frequency estimates may also be imprecise, particularly when based on few observations. Scenario comparisons clarify the consequences of assumptions, but they cannot replace functional evidence or correct a frequency that does not represent the target population.
Check Your Understanding
Answer each question using the assumptions stated in it.
- Under Hardy–Weinberg equilibrium, what is the carrier proportion when \(q=0.02\), if one copy of the allele is sufficient to count as a carrier?
- Why is an allele with unknown function not equivalent to a normal-function allele?
- A rare allele is assigned normal function in one scenario and no function in another. Name two assumptions that must be held fixed for the scenario comparison to isolate the effect of the functional assignment.
- For \(q=0.01\), calculate the homozygote proportion under Hardy–Weinberg equilibrium. Why might it still matter in a large virtual population?
Solutions. 1. The carrier proportion is \(2q-q^2=2(0.02)-(0.02)^2=0.0396\), or 3.96%. 2. “Unknown” means the evidence does not establish the functional class; assigning normal function would make an unsupported biological claim. 3. For example, hold the allele frequencies, genotype-frequency assumptions, and phenotype-classification rule fixed. 4. The homozygote proportion is \(q^2=(0.01)^2=0.0001\), or 0.01%. A large population can contain some homozygotes despite the small proportion, and the genotype may map to a distinct phenotype category.