Context of use
Name the decision, users, process range, outputs, acceptable risk, and excluded uses. A screening model for route selection needs different evidence from a model used for facility investment or GMP control strategy.
Engineering guide · Model validation
How to validate a bioprocess model for an engineering decision.
Validation begins with the decision and an agreed acceptance plan. A model can close numerically and still be unfit for scale-up, capacity, cost, sustainability, safety, or regulated use.
Write the validation plan first
Define what the model must predict and how well.
Acceptance criteria
Predefine balance closure, trajectory error, yield and titer error, equipment duty, event timing, throughput, cost and LCA tolerances. State whether criteria apply per batch, per unit, or across a validation set.
Independent evidence
Separate calibration data from validation data by batch, scale, campaign, or operating condition. Do not validate only against the same observations used to fit parameters.
Model maturity
State model maturity without ambiguity
LevelWhat is demonstratedAppropriate use
ScreeningEquations execute, balances close, assumptions and physical boundaries are visibleOption framing and data-gap discovery
CalibratedParameters fit identified data with residuals, uncertainty and validity domainScenario analysis inside the calibration envelope
ValidatedIndependent cases meet predefined acceptance criteria with documented deviationsThe stated context of use and operating envelope
Decision-readyTechnical validation, source governance, review, change control and decision risk are acceptedThe named engineering or investment decision
Normalized residualej = (ymeasured,j − ypredicted,j) / σj
Inspect residuals over time, scale, batch, operating condition, and measurement quality. A single R² value can conceal bias, autocorrelation, poor boundary behavior, parameter non-identifiability, or extrapolation outside the evidence domain.
Preserve technical traceability
Make every result reproducible and reviewable.
Source ledger
Store value, unit, low/base/high interval, source, owner, confidence, validity range, timestamp, and transformation for every material parameter.
Versioned execution
Bind inputs, code or equation version, solver settings, run time, outputs, checks, and reviewer status to one immutable run record. Compare changes before approving a new model version.
Limitations register
Expose missing data, simplified physics, uncalibrated submodels, extrapolations, failed criteria, and required follow-up work directly beside the model output.