That’s what happened to Research Fellow Michael Pan (UNSW) when he stood up at a MACSYS workshop at QUT to outline his research plans. MACSYS brings together researchers in cell biology, mathematics, statistics, machine learning and AI – and all those disciplines were represented in that room when Michael mentioned he wanted to use a method called parameter balancing for the modelling work he was doing.

“It was a genuine shock to hear that an accepted methodology in systems biology modelling was potentially based on flawed statistical assumptions that may not hold in reality,” Michael recalled.

That moment sparked an interdisciplinary effort across QUT, UNSW, Monash University, and The University of Melbourne, with researchers combining their expertise to create a new, more rigorous parameter balancing framework – one that could play a crucial role in the Centre’s mission to model an entire cell.

Kinetic models: describing the engine of the cell

To understand why parameter balancing matters, it helps to look at the models it supports. One of the most important ways to understand how cells behave is to study the enzyme-driven reactions that power metabolism. To make sense of these reactions, scientists build enzyme-kinetic models that track how fast reactions run and how changes in one part of the network ripple through the entire cell. These kinetic models supply the essential numerical information that whole-cell models rely on to simulate living systems and make predictions.

But there are two major challenges with these models.

“The data that are used come from many different experiments done in labs around the world,” explained Monash Chief Investigator Mike McDonald. “This means that, despite the best efforts from experimenters, data are incomplete and maybe inconsistent.”

The other problem, according to QUT Research Fellow TrungTin Nguyen, is that these models sometimes describe something that’s not possible.

“Some models can violate things like mass or energy conservation and produce biologically or physically impossible behaviour,” Tin said.

A new (parameter) balancing act – powered by the Bayesians

To fill in the gaps where data may be missing, or physically or thermodynamically incompatible, scientists use a technique called parameter balancing. Led by statistical researchers at QUT, the MACSYS team is taking parameter balancing to another level by incorporating Bayesian statistics.

Tin explains that the Bayesian approach considers factors like temperature or pH – key conditions that can speed up, slow down, or completely change enzyme behaviour.

“This approach fills in the gaps in the data while enforcing the laws of thermodynamics, so estimates stay chemically consistent,” Tin said. “Plus, it checks whether the model is making good predictions on data it hasn’t seen.”

From the statistics side, QUT Chief Investigator Chris Drovandi highlights two other essential ingredients.

“First of all, it quantifies uncertainty. It shows how realistic the parameter estimates are,” said Chris. “The second important element incorporates prior knowledge, starting from what we already know to estimate missing values in a biologically reasonable way. It acts as a guardrail, ensuring our parameters obey chemistry and physics.”

Together, these features turn parameter balancing into a more robust, transparent, and scientifically grounded process.

An important piece of the whole-cell model puzzle

The team sees this work as an important piece to the mission of MACSYS of creating whole-cell models with true predictive power – models that can forecast how a cell will behave under new conditions before anyone steps into a lab.

“Predictive models are essential if we want to test ideas in the computer before doing expensive, time-consuming experiments,” Mike said. “This work makes models more useful as decision-making tools rather than just explanatory diagrams.”

Tin agrees.

“From both the wet-lab and modelling sides, this work can change how confident we are in model predictions,” said Tin. “From the methodology side, we now have a calibrated, thermodynamics-aware Bayesian layer that can be shared across pathways and organisms.”

Assumptions challenged, science advances

For Deputy Director and Chief Investigator Jennifer Flegg (Melbourne), the project highlights the value of cross-disciplinary approaches.

“When people with different mindsets work together, they challenge assumptions that have gone unquestioned for years,” she said. “That’s how breakthroughs happen.”

That theme resonates with how the project began.

“This whole project started when I discussed my intentions to use parameter balancing at a MACSYS Workshop,” Michael recalled. “Working with experts in different fields forces us to question inherent assumptions that we adopt in our research.”

Tin underscored the point.

“This project needed three ingredients at once: statistical methodology and diagnostics at QUT, domain expertise and largescale reconstructions at UNSW, Melbourne, and Monash, and shared, reproducible pipelines to validate across organisms and datasets,” he said.

They’re ingredients that only an ARC Centre of Excellence can bring together.