Combining automated kinetic model selection with robust optimization and application to maximize cell-free production...

Cell-free multi-enzyme cascades enable precise biosynthesis, but using them for cost-efficient bioproduction remains challenging. This issue can be addressed through model-based process optimization using kinetic models. However, the predictive power of kinetic models is often limited by structural and parametric uncertainties. In this work, we present MoRSel, a novel method that tackles structural uncertainties through automated refinement and selection of kinetic models. MoRSel starts with a kinetic core model, a library of potential model changes, and available experimental data. The algorithm iteratively expands the model to achieve an optimal balance between fit quality and model complexity according to information-theoretic criteria. We applied MoRSel to a kinetic model of a cell-free enzyme cascade for producing the nucleotide sugar UDP-GalNAc. An initial core model showed significant inconsistencies with measured experimental data. By providing candidates for potential stoichiometric and regulatory changes in the model, MoRSel identified a model that more accurately captured the observed behavior. We then used this model to inform a recently introduced robust (ensemble-based) process optimization strategy that accounts for parametric uncertainty. We addressed two economically relevant problems. First, we determined the optimal initial substrate and enzyme concentrations to maximize the product titer, resulting in a 45 % increase in product titer and yield in a validation experiment. In a second optimization, we minimized the enzyme load and tested the computed optimal enzyme concentrations in another validation experiment. With enzyme costs reduced by 53 %, the relative titer and yield could be improved by 80 % and 75 %, respectively, with small reductions in absolute titer and yield. Overall, MoRSel advances the development of predictive models for complex biochemical systems, with particular relevance for the robust optimization of cell-free bioproduction processes.