Data-driven symbolic regression reveals microbial growth laws

Microbial growth is intimately dependent on the resources that cells find in their environment. While resource consumption in simple systems is well understood in theory, connecting mathematical models with empirical observations in more complex environments remains difficult. To address this, we measure bacterial growth across various concentrations of a complex growth medium and analyse the data with the assumption that populations consume resources at a constant rate. This approach highlights cumulative population gain as a key observable, revealing otherwise hidden resource dynamics.

We show that using cumulative population gain allows machine learning algorithms to predict growth rates more accurately than instantaneous population size, supporting the constant-rate resource consumption model. However, black-box machine learning says very little explicitly about the actual growth laws. To overcome this, we apply symbolic regression in a biologically informed way to reveal that additive and multiplicative models based on a Monod dynamics provide the best balance between data fit, model parsimony, and biological relevance.

Our study demonstrates how combining theoretical knowledge with symbolic regression can advance the understanding of microbial growth, bridging the gap between mathematical modelling and empirical research in microbial ecology.