Removing scale-up guesswork: industry wins with digital twins

Scaling up from lab success to reliable industrial production remains one of biotechnology’s biggest bottlenecks and “digital twins” are emerging as a powerful way to de-risk the process.

At UQ’s Biosustainability Hub, Dr Isabella Casini from AIBN is part of a research team developing “digital twins” to reduce the guesswork of scaling up across different bioreactor sizes.

“Digital twins could simulate large-scale conditions while an experiment is still at lab scale, to help researchers understand what cells will experience,” Dr Casini said.

“It is a ‘preview’ of industrial performance.”

This insight allows researchers to optimise bioprocess early, design more robust microbial strains and anticipate productivity losses before they occur.

The power of prediction

By pairing physical bioreactors with digital models that use real-world data and scientific knowledge, these “digital twins” could help industry predict how large-scale systems will behave.

However, a true “digital twin” of a bioreactor bioprocess does not yet appear to exist.

Current systems receive data from the bioreactor (e.g., pH, temperature) and can send instructions back to adjust these conditions.

However, these models cannot reliably predict what will happen next because they don’t capture the biology inside the bioreactor.

This makes it difficult to optimise performance in real time.

The ultimate goal is a tightly coupled system where the physical bioreactor and its digital counterpart continuously exchange information, enabling not only control but also prediction and optimisation.

“If we do this, it will be the first of its kind, because so far, digital twins have limited power to predict.” Dr Isabella Casini

In that scenario, companies could run a bioreactor alongside its digital twin, adjusting conditions in real time, reducing trial-and-error experimentation and improving consistency as processes scale up.

What is a digital twin?

A digital twin is a virtual replica of a physical system, in this case, a bioreactor, that is connected to the real system through data.

Digital twins have three parts:

• The physical system: the real bioreactor

• The digital model: a computer-based simulation of that reactor

• The data link: a two-way flow of information between the two

Sensors in the bioreactor continuously measure variables like temperature, pH and gas composition.

These measurements are fed into the digital model, which uses them to mirror what is happening inside the bioreactor in real time.

Changing the environment

At UQ’s Biosustainability Hub, researchers are using bacteria and yeast in fermentation processes (controlled microbial growth systems) to produce sustainable fuels, foods and materials.

When a fermentation process is expanded from a 1 litre flask in the lab to a 10,000 or 100,000 litre bioreactor commercially, the microbes are the same, but the environment has changed drastically, leading to different microbial behaviour and changes in productivity.

In small bioreactors in the lab, everything mixes perfectly – temperature, nutrients and oxygen are evenly distributed, conditions are well-controlled and cells behave predictably.

But on expansion, even to the pilot scale, this environment changes, leading to a drop in productivity.

Large reactors develop complex flow patterns and excess heat, mixing becomes expensive and less effective, cells can get too much or not enough nutrition.

Delicate cells can also be affected by vigorous agitation.

When detail is a bottleneck

To understand what happens inside large reactors, researchers are using computational fluid dynamics (CFD).

These are simulations that model how liquids move, how heat, gases and nutrients disperse, and how particles (representing cells) travel through the bioreactor.

CFD models create a detailed 3D ‘map’ of conditions in the bioreactor, revealing how long cells spend in different environmental “zones” of the bioreactor and these patterns can be used to see how they affect cell growth and production rates.

These detailed simulations underpin digital twins, but CFD models are highly computationally intensive.

Even for relatively small systems, simulations can take days to run.

That makes them too slow to use in real-time decision-making.

Enter the surrogate

Because CFD simulations are too slow for real-time decision-making, the team is developing surrogate models – fast, machine learning-based versions trained to approximate the results of full CFD simulations.

“Once trained, a surrogate model can produce predictions in milliseconds or seconds rather than days,” Dr Casini said.

The models are trained and validated using wet-lab data from bioreactors at the UQ Biosustainability Hub, generated by team member Dr Ian (Alex) Petersen.

That speed could allow a digital twin to combine real-time bioreactor data with biological models, predict conditions inside the bioreactor, and recommend changes to operating conditions before problems arise.

Ideally, those adjustments – such as to mixing speed, nutrient feeds, pH or temperature – could be tested virtually before being applied in the real system.

The long-term vision

If the long-term vision of the digital twins is realised, this approach could significantly reduce one of biotechnology’s biggest bottlenecks – turning promising lab results into reliable, large-scale production.

This project is funded by TJX Bioengineering and FaBA.

Dr Casini is working with industry partner TJX Bioengineering, makers of bioreactors and has received a Marie Curie Postdoctoral Fellowship focusing on bioprocess scale-up (anticipated start 09/2027).

The team also includes Dr Tim McCubbin, Dr Axa Gonzalez and Professor Esteban Marcellin.

UQ's Biosustainability Hub uses synthetic biology to help the world’s biggest businesses transition to net zero.
Funded by government, industry and UQ, the $70 million Biosustainability Hub is a one-stop-shop for big companies to transform their production practices and create carbon neutral economically viable products and materials.