Bayes Reading Group: Nicholas Cowie (Technical University of Denmark)

Friday Apr 11, 2025, 11:00 AM → 12:00 PM America/New_York

3rd Floor Conference Room (162 Fifth Avenue)

Discussion Lead: Nicholas Cowie (Technical University of Denmark)

Topic: Non-parametric models of cell metabolism

Abstract: Metabolic networks are often modelled as systems of ordinary differential equations with the form dx/dt = f(x, e, θ), where x are metabolite concentrations, e are enzyme concentrations and θ is the set of hyperparameters. The previous gold standard for non-parametric metabolic models uses a single reference experiment to calculate dx/dt and is only locally accurate. Predictions about metabolite concentrations and reaction rates are therefore highly influenced by the choice of the reference experiment. We propose a reproducing kernel Hilbert space framework that uses a set of inducing experiments – eliminating the need to identify the best reference condition and creating a more accurate representation of metabolism. This presentation evaluates how well this approach works when modelling a small metabolic pathway by comparing predictions made by this method to the previous gold standard. Furthermore, we will show how this method better approximates aspects of enzyme behaviour that break the assumptions underlying standard parametric models.