A practical audit framework for deciding whether available physicochemical, in vitro, in vivo and clinical evidence is sufficient for the model's intended use.
Assess Your PBPK Data ReadinessA PBPK project is ready when the available data can identify and test the biological processes that control the decision-relevant outputs—not when every possible parameter field has been populated. Minimum data may support an exploratory model for hypothesis generation or study design. A stronger, internally consistent dataset may support quantitative prediction across doses, routes or drug-interaction scenarios. Evidence intended to influence a regulatory or high-consequence clinical decision requires a higher standard: traceable parameter provenance, qualified system components for the intended use, biologically plausible model structure, independent or appropriately separated verification data, prespecified performance criteria, sensitivity and uncertainty analyses, and reproducible reporting. Therefore, minimum data ≠ ideal data ≠ regulatory-support evidence. The same compound can be "ready" for an early exposure scenario but not ready to replace a clinical study. Begin by defining the context of use, then map each influential model mechanism to measured evidence, remaining assumptions and a plan for reducing uncertainty.
Enough information to construct an exploratory model and expose key gaps.
Measured, mutually consistent inputs and relevant PK data support a defined development decision.
The evidentiary package is proportional to the intended regulatory impact.
"Regulatory-support" does not mean automatic acceptance. FDA states that using PBPK results in lieu of clinical PK data is assessed case by case according to intended use and the quality, relevance and reliability of the analysis. EMA similarly links the required depth of qualification and model evaluation to how much weight the simulation carries in the decision.
Use the exact molecular form, salt, stereochemistry and formulation relevant to the simulated route. Measured values are preferred when an input is influential and prediction uncertainty is large.
Assay system, species, matrix, concentration, recovery and binding corrections must accompany the numerical value. The appropriate set depends on whether absorption, metabolism, transport or interaction is central to the question.
IV and oral PK in informative species can separate clearance, distribution and bioavailability. Tissue data, mass balance or metabolite profiles may be needed when the intended model claim depends on a specific organ or pathway.
Available human PK should cover conditions that challenge the components used in the prediction. A single profile can be useful for calibration but may not verify dose proportionality, pathway contribution or population extrapolation.
Record whether a value was measured, literature-derived, predicted, optimized or taken from a platform default. Include method, units, biological system, molecular form, date and reference.
Separate biological variability from uncertainty about a parameter value or mechanism. Provide range, distribution or confidence information when available, rather than a precision-free point estimate.
Explain conflicts between sources, binding corrections, scaling factors, fitted values and any transformation into the parameter used by the model. Plausibility must survive the conversion.
Parameter provenance is not an administrative appendix. It determines whether a sensitivity result can be interpreted and whether another analyst can reproduce the same simulation. If several credible values exist, selecting the one that best fits a profile without documenting alternatives can hide structural uncertainty.
The required evidence should be scaled to the intended use. "Minimum" below means sufficient to begin a transparent exploratory analysis, not a universal threshold for every model.
| Evidence domain | Minimum Exploratory model | Ideal Development decision | Regulatory support Higher-impact use | Readiness question |
|---|---|---|---|---|
| Context of use | Compound, route, population and question stated | Decision, alternative actions and output precision defined | Claim, regulatory purpose and consequences of error documented | Would the simulation change a study, dose, label or restriction? |
| Physicochemical / product | Identity plus plausible pKa, lipophilicity, solubility and binding | Measured influential properties under relevant conditions; formulation represented | Validated methods, traceable lots/forms and uncertainty ranges; product inputs appropriate to claim | Could a different solid form, pH condition or free fraction change the outcome? |
| In vitro ADME | Major clearance and absorption hypotheses represented | Quantitative human systems with recovery, binding and pathway characterization | Fit-for-purpose assays, controls and cross-system consistency for mechanisms supporting the claim | Are important pathways measured, or merely assigned to a residual term? |
| Preclinical in vivo | At least one informative PK dataset for plausibility | IV/oral, dose and formulation conditions that test CL, distribution and F | Datasets selected to test translatable mechanisms, with discrepancies investigated | Does the species express the mechanism in a way relevant to humans? |
| Clinical PK | Not always available for an early exploratory model | Human profiles across informative dose or regimen conditions | Independent or prespecified data that challenge the model components used for extrapolation | Does the dataset verify the intended mechanism or only the overall curve? |
| Model evaluation | Visual predictive checks and broad sensitivity analysis | Prespecified metrics, uncertainty analysis and alternative structural hypotheses | Fit-for-purpose platform support, drug-model evaluation and impact-aligned acceptance criteria | Were criteria defined before the prediction was judged? |
| Reproducibility | Inputs, equations and settings recorded | Versioned files, executable workflow and simulation records | Complete report, audit trail, software/platform version and reproducible tables/figures | Can an independent analyst recreate the result and understand every change? |
Parameters are adjusted using observed data so that simulated results describe those same observations. This can be scientifically appropriate when the adjusted parameter is identifiable and biologically plausible.
The model is challenged with data, conditions or mechanisms not used for the relevant calibration step, selected because they test the intended application.
Data separation need not be mechanically absolute in every development setting, but its limitations must be explicit. When all available human data are used for optimization, there may be no independent evidence left to establish predictive performance. A later study can then be planned to challenge the model. Verification should focus on the mechanism supporting the intended extrapolation: matching AUC alone may be inadequate if the decision depends on Cmax, half-life, intestinal first-pass loss or a specific clearance pathway.
Sensitivity, variability and uncertainty are related but should not be reported as synonyms.
A highly sensitive parameter is not automatically uncertain, and an uncertain parameter is not necessarily influential. The combination matters. A measured but highly variable absorption parameter may require population simulation; a poorly known but insensitive tissue partition coefficient may not justify another experiment. Readiness improves when the team can explain which missing data would materially narrow the prediction and which would not.
Available inputs, formats, assay context, quality observations and ownership.
Proposed absorption, distribution and clearance structure with evidence strength.
Missing or uncertain parameters ranked by likely impact on the intended output.
Calibration, verification, sensitivity, scenarios, acceptance criteria and reproducible deliverables.
The result may be "ready to build," "ready for exploratory use," or "additional data needed before the intended claim." The last outcome is useful when it identifies a targeted experiment rather than requesting a generic package of every possible assay.
Practical principle: build the smallest model that represents the mechanisms required by the question, but document it to the standard demanded by the decision.
A PBPK engagement can begin with intended-use definition, input-data audit, mechanism and parameter mapping, gap prioritization, model development, verification planning, sensitivity analysis and reproducible reporting. Scope should be matched to the evidence available and the decision to be supported.
Explore PBPK Modeling and Simulation →Submit your project details below, and our team will respond within 24 hours.
Talk to our technical team about your project!
I Want To Talk
