Speaker
Description
Microbial communities play key roles in biotechnology, environmental systems, and human health. Understanding and predicting the dynamics of interacting microbial populations is essential for both advancing fundamental knowledge and enabling the rational design and optimization of these systems. Mechanistic models provide a powerful framework for this purpose, enabling hypothesis testing and predictive simulation and control. However, formulating such models remains challenging due to biological complexity, limited experimental data, and the number of possible interaction mechanisms.
We present a framework to automate the formulation of dynamic mechanistic models. The approach considers multiple formulations compatible with the observed microbial community behavior enabling data fitting and discrimination among candidate structures by solving a mixed-integer dynamic optimization (MIDO) problem.
The framework is illustrated through a case study involving a microbial community of three interacting species under realistic time-resolved experimental data. A grammar of mechanisms describing bacterial growth, inhibition, and substrate consumption is used to generate alternative model structures. The MIDO problem is solved using a self-adaptive cooperative multimethod algorithm. Results show the method successfully recovers the true system even in the presence of experimental noise, highlighting its potential as a scalable tool for automated model discovery in complex systems.
