From Star Alleles To Actionable Genotypes

From Molecular Calls to Analysis Groups

Pharmacogenomic data often begin as variant calls, but an association analysis usually needs a different variable: a clearly defined group that represents a gene-level genotype or predicted function. The translation matters. The same molecular call can be assigned differently if the allele definitions, copy-number interpretation, function table, or phenotype mapping rule differs. A defensible analysis therefore records each step, rather than treating a label such as “normal metabolizer” as if it were a raw laboratory measurement.

The aim here is to define an analysis-ready genotype variable, not to determine a prescribing recommendation. A predicted phenotype is a functional classification derived from genotype using specified rules; a drug-specific recommendation is a separate clinical decision. Even when a guideline uses the phenotype to inform treatment, the analysis must state how the phenotype was obtained and which groups it compares.

These are related but not interchangeable objects. A star-allele label summarizes a molecular haplotype; a diplotype combines haplotypes; a phenotype interprets the genotype’s expected function; and an analysis group is chosen for a particular question. The group may retain phenotype categories separately or combine them according to a prespecified scientific rationale. The translation should not conceal uncertainty in the underlying call.

Specify the Translation Rules Before Grouping

A reproducible translation requires a defined reference for each stage. At minimum, record the genome build and variant-calling procedure, the star-allele definition source and version, the method used to determine phase, the copy-number assay and interpretation rules, and the function annotation and phenotype mapping version. For a multi-gene panel, these rules may differ by gene.

The interpretation also relies on assumptions that should be checked rather than left implicit:

• The assay measures the variants needed to distinguish the relevant star alleles, or explicitly reports when it cannot.
• The allele definitions and functional annotations apply to the reference and transcript used in the study.
• Phase is known or its uncertainty is represented; variants observed in one person do not automatically reveal which haplotype carries each variant.
• Copy number is measured or inferred with a method appropriate to the gene, and the number of copies is not confused with the number of distinct allele labels.
• The phenotype mapping is the stated gene-specific rule. It is not assumed to be universal across genes, guideline versions, or drugs.

A common approach for genes with an activity-score system is to assign each allele a function value, combine the values across copies according to the gene-specific rule, and map the resulting score to a phenotype category. This is a useful framework, not a universal law: some genes use different functional classifications, and some allele effects cannot be summarized by a simple additive score.

Worked Translation With a Functional Score

The following examples use deliberately hypothetical rules. They illustrate the translation procedure and are not clinical guidance or a phenotype map for a real gene.

The score calculation has a simple justification under its assumptions. If each allele copy contributes its assigned function value and the gene-specific model defines total function as the sum of copy contributions, then the total is obtained by adding the contribution of every copy. This explains why copy number must be handled before assigning a score. It does not establish that additivity is biologically appropriate for every pharmacogene.

Here \(C\) is the number of gene copies represented in the call and \(a_j\) is the assigned functional value of copy \(j\). This expression applies only when the selected gene-specific scoring system uses that additive rule. If a gene’s established interpretation uses a different rule, use that rule and document it instead.

Copy Number and Allele Assignment

A duplication or deletion can change the function implied by a diplotype label. In a copy-number-aware call, it is not enough to know that a duplication exists: the interpretation may depend on which allele was duplicated. A notation such as \(*1x2/*4\) indicates two copies of \(*1\) and one copy of \(*4\), whereas \(*1/*4x2\) indicates a different distribution of allele copies. Those calls can produce different predicted function under the same allele-value table.

When copy number is uncertain, or the duplicated allele cannot be assigned, do not choose whichever interpretation produces the more convenient phenotype. Record the unresolved copy-number state or allele assignment, and determine in advance whether the analysis will classify that observation as indeterminate, use a prespecified set of compatible interpretations, or exclude it from a particular analysis. The choice and resulting denominator should be reported.

From Phenotype Categories to Model Variables

After translation, the analysis group must match the scientific question. A categorical predictor with \(K\) levels can be represented with indicator variables. If one level is chosen as the reference, the remaining \(K-1\) indicators compare each corresponding group with that reference. This encoding makes the model’s contrasts explicit; it does not change the underlying genotype classification.

For example, a study might retain poor, reduced, and typical function as three groups and specify typical function as the reference. Alternatively, it might define one clinically motivated contrast in advance. Either way, state which translated categories enter each group and which group is the reference. A label such as “actionable genotype” is not an adequate variable definition unless its membership rule is given.

Document Ambiguity and Preserve the Audit Trail

Ambiguity can arise at several stages: a variant may be unmeasured, a star-allele definition may match more than one allele, phase may be unknown, or copy number may be uncertain. These are not interchangeable problems. A useful translation record identifies the stage and the reason for uncertainty, so a later analyst can distinguish “not tested” from “tested but not resolved.”

The distinction between “unknown” and “no function” is especially important. An unresolved genotype is not evidence of a null-function genotype. Likewise, a function category derived from a genotype does not prove a drug response for an individual. The translated variable represents the classification rule used in the analysis, while observed response remains an outcome measured under a particular treatment context.

Common Mistake: Treating the Phenotype Label as Self-Explanatory

A tempting but incorrect approach is to replace every molecular call with a phenotype label and discard the supporting information. The reasoning is that the label already summarizes the genotype, so the details are unnecessary. This can hide whether copy number was considered, whether phase was resolved, and which annotation version produced the label. It also makes reclassification difficult if a rule changes.

The correction is to retain the molecular call and the derived phenotype as separate variables, document the translation rules, and preserve uncertainty explicitly. Analysis groups should then be defined from those documented phenotypes. This keeps the statistical contrast interpretable and the classification auditable without implying that the group itself is a clinical recommendation.