Data-Driven And Mechanistic Modeling For The Optogenetic Control Of Microbial Consortia

Abstract

Metabolic engineering offers a sustainable way to manufacture bulk and specialty chemicals through microbialfermentation of renewable feedstocks. However, maintaining high productivity is a challenge, primarily due to the lack of tools to control cellular metabolism. Microbial optogenetics offers a solution by enabling the dynamic and reversible modulation of metabolism using light as an input. With the possibility for any gene to be put under light-control, and for a time-varying light input to dynamically control gene expression, this represents a huge parameter space. Hence, there is a need for high-throughput experiments and for data-driven and mechanistic models to drive understanding, optimization, and control of the biological systems. This thesis outlines the development of novel optogenetic hardware and software that facilitates bothhigh-throughput experiments and automated closed-loop regulation using fluorescent biosensor outputs. For data-driven learning from these experimental data, machine learning algorithms are developed that enable learning from data with various 'pathologies' often encountered in experimental systems without the need for data modification. These include data collected at variable time intervals, and data with only partial state-space observations at each point. To demonstrate how the model can leverage not just experimental data but also a priori knowledge, the data-driven models are merged with existing partial physical insights in "physics-informed gray-boxes" to enable learning of understandable functions like unknown kinetic rates and microbial growth functions. Optogenetics can permit the multi-modal regulation of microbial growth rates using a single light wavelength. By employing forcing periods and pulsing fractions as dual and independent parameters of optical input, it is possible to manage both the relative growth rates of two species grown in consortia and their absolute growth rates. Finally, closed-loop control over microbial growth rates is demonstrated in continuous cultures and the extension to the control of population ratios in microbial consortia is discussed.