Not every scale-up decision needs a computational model. Some questions are best answered by running the bench experiment. Some require only a quick engineering calculation on the back of a spreadsheet. Some genuinely benefit from a full FBA-based flux simulation. Getting confused about which is which costs time and money in either direction — over-modeling simple decisions wastes effort, and under-modeling complex ones causes failed runs.
This article offers a practical decision framework for fermentation scientists working through scale-up, based on the nature of the uncertainty and the cost of getting it wrong.
Category 1: Run the Bench Experiment
Some questions about your process are best answered empirically because the parameter space is small and the experiment is fast and inexpensive.
Appropriate for bench experiment:
- Optimal temperature and pH for your specific strain and construct. These are gene expression and enzyme kinetics questions. FBA cannot predict them from first principles without extensive parameterization. Screen them empirically at 2L or in microbioreactors.
- Media component concentration optima (trace element concentrations, nitrogen source choice, phosphate levels). These affect growth kinetics in ways that are not well-captured by stoichiometric modeling. Run a DOE (design of experiments) screen.
- Induction condition optimization (IPTG concentration, temperature-shift magnitude, inducer timing in fed-batch). These require empirical characterization of the expression system. Model outputs will be less informative than a well-designed bench experiment.
- Anti-foam selection and concentration. Anti-foam effect on kLa is highly product-specific and difficult to model. Measure it directly in your media.
The decision rule: if the answer depends on biology that is specific to your construct or strain and is not represented in your stoichiometric model, run the bench experiment. If the answer requires understanding physical parameters that change predictably with vessel scale, consider modeling.
Category 2: A Spreadsheet Calculation Is Sufficient
Many scale-up questions can be answered with simple engineering calculations without constructing a full FBA model. These are questions about the physical environment of the vessel, not about metabolic flux distributions.
Appropriate for engineering calculation:
- Minimum agitation rate to maintain DO above setpoint at target OUR. If you know OUR from off-gas analysis and kLa from empirical correlations, the minimum agitation rate is directly calculable from the kLa-agitation relationship for your impeller type. No FBA required.
- Maximum feed rate given a target DO setpoint and known kLa. Feed rate ceiling = (kLa × DO_driving_force) / (yield coefficient × specific oxygen demand). This is a single calculation, not a simulation.
- Mixing time at target operating conditions. Mixing time scales as tm ≈ A × (V/P)^⅓ for standard geometries, where P is power input. Published correlations for common impeller types give you this directly.
- Hydrostatic DO correction at commercial scale. The correction factor for dissolved oxygen partial pressure as a function of vessel depth is a straightforward calculation from vessel liquid height and operating pressure.
The decision rule: if the question involves one or two physical parameters with established correlations, calculate it. Building a simulation infrastructure for it is over-engineering.
Category 3: Model It — Simple FBA
Simple FBA is appropriate when you need to understand how a change in one or two exchange flux constraints changes the metabolic state of your culture — specifically, whether the culture will operate in overflow or non-overflow mode under the target conditions.
Appropriate for simple FBA:
- Overflow risk at pilot scale given bench-characterized phenotype. Set oxygen uptake bound to the pilot vessel kLa limit. Run FBA. Does the model predict acetate secretion (E. coli), ethanol secretion (yeast), or fumarate secretion (F. venenatum)? If yes, by how much, and at what feed rate can it be prevented?
- Titer prediction under oxygen-limited conditions. If you know kLa is insufficient to maintain full aerobic metabolism, what yield can you expect under partially limited conditions? FBA with a modified oxygen bound gives you this estimate.
- Feed strategy comparison. Run FBA at three feed rate profiles (exponential, linear, DO-stat). Which keeps the exchange fluxes in the non-overflow regime longest? This is a constrained optimization question that FBA answers directly.
- Strain variant comparison before pilot. You have three candidate strains with different bench-scale flux phenotypes (different qs, μ, qp). Which one will maintain the best productivity under the oxygen-limited conditions predicted at pilot scale? FBA gives you a rank order that bench scale cannot.
Category 4: Wait — You Need More Bench Data First
There are situations where a model will give you a precise answer to the wrong question — because the input data is insufficient to constrain the model usefully.
Wait for more data when:
- You have only batch data, no fed-batch data, and the scale-up is for a fed-batch process. FBA constrained by batch exchange fluxes will not accurately predict overflow behavior in a fed-batch where glucose is artificially limited and cell density is much higher. Run at least one fed-batch bench run before building the scale-up model.
- Your titer measurements are inconsistent across bench replicates (>30% CV). The model's productivity predictions will be dominated by measurement noise. Fix your assay reproducibility first.
- You haven't measured acetate (or your relevant overflow metabolite) during bench fed-batch. If you don't know what the overflow metabolites are doing at bench scale, you can't validate whether the FBA model's overflow predictions are calibrated to your strain.
- Your strain has significant genetic modifications that change overflow pathways. Standard FBA models for E. coli, yeast, or F. venenatum are based on wild-type or minimally modified metabolic networks. If your strain has deleted or overexpressed pathway enzymes that affect overflow metabolism, the standard model will give misleading overflow predictions. Validate with bench experiments first.
The Decision Heuristic
As a working heuristic, apply the following sequence to any scale-up process development question:
- Is the parameter determined by biology specific to my strain or construct (temperature optimum, expression timing, media components)? → Bench experiment
- Is the parameter determined by physical vessel engineering with established correlations (kLa, mixing time, hydrostatic pressure)? → Engineering calculation
- Do I need to understand how a physical constraint at the target scale changes the metabolic state of my culture (overflow risk, titer under O₂ limitation, feed ceiling)? → FBA simulation
- Do I lack the bench-scale flux data to constrain a useful FBA model? → Wait, run more bench experiments first
The framework is not about which tool is more sophisticated. FBA is not "better" than bench experiments — they answer different questions. The engineer who runs a bench screen for temperature optimization and then applies a 30-minute FBA calculation to predict overflow risk at 500L is using both tools correctly. The engineer who builds a full genome-scale metabolic model to answer "what's the optimal IPTG concentration?" is using a complex tool to answer a question that a $20 bench DOE would resolve faster and more reliably.
Boundary Statement: When Not to Model
Fermvyne's FBA-based scale-up prediction is designed for aerobic fed-batch and continuous processes where oxygen transfer is a limiting factor at scale. It is not designed to replace bench characterization of your specific strain's biology, to predict the effect of genetic modifications without experimental validation data, or to substitute for hands-on kLa measurement at your target vessel. Use it for the questions it's built to answer: what changes physically at the target scale, and what does that mean for your flux distribution and titer?
References
- Noorman HJ. An industrial perspective on bioreactor scale-down: what we can learn from combined large-scale bioprocess and model fluid studies. Biotechnol J. 2011;6(8):934–943.
- Doran PM. Bioprocess Engineering Principles. 2nd ed. Academic Press; 2013.
- Schmidt FR. Optimization and scale up of industrial fermentation processes. Appl Microbiol Biotechnol. 2005;68(4):425–435.