Machine learning uncovers analytical kinetic models of bioprocesses

Identifying suitable kinetic models for bioprocesses is a complex task, particularly when interpretable models are sought. Classical machine learning algorithms are gaining wide interest to simulate complex bioprocesses that are hard to describe via first principles. However, they often rely on a priori assumptions of the model structure and lead to mathematical expressions that are hard to interpret. In this work, we apply an alternative approach based on symbolic regression to identify bioprocess models without assuming a pre-defined model structure. We obtain algebraic expressions for the kinetic rates from data consisting of concentration profiles. The model training was performed following a two-step approach that allows avoiding the iterative integration of differential equations for the parameter estimation step. The proposed procedure was found from numerical examples to slightly outperform neural network benchmarks. Moreover, the obtained algebraic expressions for the rate equations facilitate the model interpretation and enable the direct application of optimization algorithms.