
Bioprocessing does not have a shortage of ambitious data models.
This is one of the learnings from the 2026 Bioprocessing Summit event held in Boston, which CLADE attended in August.
Digital twins appeared in discussions about UF/DF, viral vectors, media optimisation, production scale-up and continuous process control.
The key benefits of digital twins in biopharma are easy to understand: we can test more scenarios in silico, anticipate deviations and make better process decisions.
Every one of the models, however, depends on quality data collected on real samples at specific points in a process, within a specific context.
So before asking what a digital twin can predict for your bioprocess, we should look at a practical question: what analytical data is available and what has it proved?
What data and process decisions are needed for a bioprocess digital twin
For many bioprocessing teams, protein concentration is not the only analytical attribute they need to monitor in samples. They want to see it alongside buffer or excipients data and connect those results to the same sample and process point.
The analytical data including detailed context should be recorded and ideally, consolidated so that it’s ready to feed the models.
In terms of the process, it is crucial to answer this question:
At [x] point in the process, what will we do differently when new analytical insights are available?
For example, during UF/DF, the process decision required might be whether diafiltration should continue or not.
Once you have the clarity on the process decision required, the study becomes much tighter. It defines which samples matter, where they should be taken, which independent analytical method should be used for comparison and what difference would be large enough to change the decision.
What the analytical data needs to show for solid data models
Analytical performance in the actual matrix and range
Good performance of an analytical method in neat standards or a simple buffer may not be the same as performance in a protein-rich process sample.
Proteins, buffers, excipients, surfactants and pH can influence a spectroscopic measurement individually and in combination. The evaluation should therefore cover the formulations, concentration range and process conditions in which the method is expected to operate.
Boundary samples are particularly useful. They show where performance begins to deviate and help distinguish a defined operating range from a general platform claim.
The aim is not to test every conceivable formulation. It is to test the part of the application space that matters for the intended decision.
The complete bioanalytical workflow
Analytical performance includes everything that happens between taking the sample and reviewing the result.
Sampling location, hold time, mixing, direct or diluted measurement, loading, rinsing and carryover can all affect the outcome.
For high-concentration samples, direct and diluted workflows should be evaluated as distinct options unless their equivalence has been shown. Sample volume, recovery, cleaning effort and time to result are part of the method’s practical performance.
Agreement with an independent analytical method for reference
Where possible, samples should be measured using an appropriate independent analytical method as well.
However, a feasibility study does not need to become a full validation exercise. It can still be rigorous if the question, comparator, sample set and acceptance criteria are clear from the start.
Analytical performance on new samples
The samples used to evaluate a model should not all be the same samples used to build and adjust it.
Held-out batches, formulations or concentration levels provide a more realistic view of how the method may behave on the next sample. Closely related replicates can otherwise produce an optimistic picture of performance.
A recent study on transferring a Raman model between cell lines illustrates the point. The model transfer was successful, but analytical performance was still influenced by correlations within the source process. Process knowledge and representative samples remained essential.
For an analytical platform, accuracy on familiar samples is only part of the evidence. The more valuable result is a clear understanding of what will transfer and what still needs product-specific verification.
A stated boundary and lifecycle
Platform does not mean that it’s universal for multiple use cases or different bioprocesses.
It means having a sound starting point, a defined area of applicability and a practical route for verification and change control. That distinction matters to biopharma companies and CDMOs that want to reuse analytical work across products, teams, instruments or sites.
The new PDA/ANSI Standard 07-2026 addresses analytical procedure transfer, comparability and platform analytical procedures through scientifically designed, risk-based studies. A recent industry and regulator workshop report reaches a similar conclusion: the boundaries, applicability and lifecycle of a platform approach need to be explicit.
Experimental context is part of the result
Analytical data becomes reusable when its context travels with it.
Another scientist should be able to identify the sample, process point, handling steps, instrument, analytical method and model version behind a result. The raw spectrum, initial quality assessment, reported endpoint and any exclusions should remain connected.
That information helps during investigations, comparability studies and method transfer. It becomes even more important when analytical results are used to build or monitor a process model.
The joint FDA and EMA Good AI Practice principles put context of use, data governance, documentation, performance assessment and lifecycle management at the centre of good practice. These considerations begin well before an AI model is deployed.
They are also relevant when no digital twin is planned yet. A result that another scientist can understand, challenge and reproduce is already more useful to Process Development, QC and MSAT.
From sample measurement to a usable result
At CLADE, we designed a connected analytical workflow:
The MIRA Analyzer captures a mid-IR fingerprint of the sample in a four-minute run. The Workbench data analytics tool helps scientists inspect spectral quality and investigate anomalies before further processing. Sphere, the data analytics platform provides the environment for applying analytical methods, while the QA Scan tool turns FTIR spectra into endpoint results for selected quality attributes.
No individual step makes an application universally transferable. The value comes from connecting measurement, quality assessment, method and result within a clearly defined use case.
That is also the foundation a useful digital twin needs.
When evaluating a new at-line application, begin with the process decision, the expected operating range and the samples that make the measurement difficult. Build the study around those conditions, pair it with an independent reference and make the method’s limits visible.
A digital twin does not begin with a model. It begins with a measurement another scientist can understand, challenge and use.
Evaluating an at-line application in formulation, UF/DF, buffer QC or release testing?
Talk to CLADE about a focused feasibility study built around your process decision and samples.
Author

Dominik Sievert
CEOContact CLADE
Connect with CLADE
Subscribe to the newsletter
Inquire about a product
Request a quote, send us a sample for analysis or other inquiries.
Talk to an
expert
Schedule a call with a CLADE expert and discuss your needs.
Follow CLADE on LinkedIn
Join this community to keep updated on latest developments.



