Building a Virtual Population With Pharmacogenotypes

From a Target Population to a Cohort Specification

A pharmacogenetic PBPK simulation should represent a clearly defined target population. That requires more than assigning a genotype to each virtual subject: the specification must say which population is being represented, how its genetic categories are distributed, how genotype and relevant modifiers map to functional phenotypes, and what physiological variability is intended. These are cohort design choices. They describe the population before any particular subject’s parameter values are set.

The earlier material on genetic variation, diplotypes, phenotype frequencies, and genotype-dependent PBPK parameters provides the necessary building blocks. Here the task is to assemble those building blocks into a coherent population-level specification. The aim is not to describe a particular sampling algorithm, but to define what the resulting cohort is intended to represent.

Specify the Target Before Combining Characteristics

Begin by stating the target population in operational terms. For example, the target might be adults with a specified disease, receiving a particular concomitant treatment, within a defined age range. A population can then be divided into mutually exclusive and collectively exhaustive strata when strata are needed to represent meaningful differences, such as age bands or population subgroups.

Let \(k\) index target strata, and let \(w_k\) be the proportion of the target population in stratum \(k\). These weights must satisfy \(w_k \geq 0\) and \(\sum_k w_k=1\). Let \(g\) index mutually exclusive genotype categories, and let \(q_{g\mid k}\) be the frequency of category \(g\) within stratum \(k\). Within each stratum, the genotype frequencies must also be nonnegative and sum to one. These definitions describe a target distribution; they do not say how genotypes will be assigned to a finite set of virtual subjects.

This is the law of total probability: the target frequency of a genotype category is the sum of its within-stratum frequencies, each weighted by the proportion of the target in that stratum. It applies when the strata partition the target and the conditional frequencies refer to those same strata. A frequency taken from one source population should not be treated as \(q_{g\mid k}\) for a different target stratum without justification.

The calculation is valid only if the two strata are defined consistently and the weights describe their shares of the intended target. If the strata overlap, omit part of the target, or use weights from a different population, the weighted result does not describe the stated target. Genotype categories should also be defined so each person belongs to exactly one category; otherwise category frequencies can be double-counted.

Make Phenotype Rules Explicit

A genotype category and a functional phenotype are related but are not interchangeable. A phenotype rule states how genotype information is translated into a model-relevant functional category. In some applications that rule is based on a genotype-derived activity score; in others it includes a modifier such as a concomitant inhibitor. State the rule explicitly, including its thresholds, modifier values, and the PBPK parameter affected. A phenotype label should not be used as a substitute for documenting the underlying rule.

For clarity, let \(A_g\) denote a hypothetical activity score assigned to genotype category \(g\), and let \(m\) denote a modifier applied to that activity in the model. One possible illustrative rule is to calculate \(A_{\mathrm{eff}}=A_gm\), then classify effective activity below 0.25 as poor, activity from 0.25 up to but not including 1 as intermediate, and activity of 1 or greater as normal. Those cut points are part of this hypothetical model specification, not universal clinical thresholds. The cohort specification should distinguish the genotype-derived value from the modified effective value.

A rule may instead define phenotype solely from genotype, or may include several modifiers. Whatever the choice, avoid assigning a phenotype label first and then selecting parameter values that merely seem to fit it. Define the mapping, and then derive the model-relevant category and parameter consequences from that mapping. If phenotype categories are used to report results, confirm that the categories cover all allowed effective values and that boundary cases have unambiguous assignments.

Represent Physiological Variability as a Population Property

Physiological characteristics such as body weight, organ size, blood flow, and enzyme abundance can vary among members of a target population. A specification may describe these characteristics using distributions, conditional distributions, or carefully defined strata. A distribution is a statement about the target population, not a value for any one virtual subject. For example, a target mean body weight of 74.8 kg does not mean that every subject should have body weight 74.8 kg.

The intended joint structure matters. If genotype frequencies differ by stratum, the genotype distribution should be conditional on stratum rather than treated as a single universal frequency. Likewise, if physiological characteristics are intended to differ across strata or genotype categories, specify those conditional distributions or relationships. A set of separate marginal distributions alone does not necessarily define how characteristics co-occur in individuals. This tutorial’s specification should therefore state the intended dependence where it is relevant, without mistaking a marginal summary for a complete joint population description.

Build and Audit a Coherent Specification

A practical specification can be assembled in a sequence that keeps population-level choices visible. The steps below define the target and its intended structure; they do not prescribe a method for generating a finite cohort.

These checks help identify design failures before PBPK simulations are interpreted. A stratum weight that does not sum with the others to one may signal an omitted or overlapping stratum. Frequencies that sum to more than one may indicate overlapping genotype categories or incompatible denominators. A phenotype mapping that leaves a value at a threshold unclassified is incomplete. A parameter multiplier applied to a total clearance when it was intended for one pathway-specific intrinsic clearance can create a mechanistic inconsistency even if the arithmetic is correct.

A further diagnostic is to trace each individual-level parameter back to its population-level rule. For every genotype-dependent parameter, ask which genotype category determines it, which modifiers alter it, and whether the parameter units and reference values match. For every physiological value, ask which target distribution or conditional distribution it is meant to represent. If that trail cannot be stated clearly, the cohort design is not yet reproducible.

Choosing Among Cohort Designs

A population-weighted design represents the target composition through its stratum and genotype proportions. A stratified design deliberately represents specified subgroups, which can be useful when subgroup-specific behavior is a primary question. A representative-subject design uses a small number of typical parameter sets, which can be efficient for illustrating mechanisms but may not represent population variability or tail behavior. These approaches answer different questions; none is automatically appropriate for every analysis.

The design should be selected in light of the prediction question. If the goal is a target-population exposure distribution, replacing the target composition with equal numbers in each genotype category changes the represented population unless that equal allocation is explicitly corrected in the analysis. If the goal is to compare genotype-specific mechanisms, deliberate subgroup representation may be useful, but the design should not be described as a natural-prevalence population unless it has that composition.

Common Mistakes and Limitations

• Treating a target frequency as a fixed cohort count. A frequency of 0.29 implies an expected proportion of 29%, not an exact count in every finite cohort.
• Combining incompatible sources. A genotype frequency from one population cannot be substituted for another stratum’s frequency without a defensible basis.
• Confusing genotype with phenotype. Genotype is a genetic category; the model’s phenotype may also depend on an explicit modifier rule.
• Confusing a population summary with a subject value. A mean or distribution describes a population; it does not assign a value to each individual.
• Assuming marginal distributions define the joint population. Separate summaries do not establish how physiological characteristics and genotype co-occur.

Every virtual population is an approximation to a target, and its quality depends on the quality and applicability of its inputs. Sparse data may leave some genotype frequencies or physiological relationships uncertain. Simplified phenotype thresholds may be useful for a model but should not be presented as universal clinical classifications. Sensitivity to plausible alternative specifications is often appropriate when important inputs are uncertain. The final specification should make those choices and limitations visible so that predictions can be interpreted in the population and mechanistic context they actually represent.