Adaptive Model Predictive Control of an optogenetic system

In vivo application of model predictive control to biochemical networks poses several challenges: the controlled system is often nonlinear and poorly known; it is susceptible to disturbances from the intra- and extracellular environment; and its dynamical behavior can vary from day to day, or even during the course of an experiment. These features can lead to a dramatic deterioration in the tracking performance of the typical Model Predictive Control (MPC) scheme. Working with in vivo feedback control of an optogenetic system in E. coli, we developed an adaptive MPC scheme capable to overcome these problems, based on the Marginal Particle Filter for joint online state, parameter and disturbance estimation. By continuously updating the model information based on the observed behavior of the controlled system and taking into account uncertainty in model behavior over the prediction horizon, this adaptive MPC scheme operated reliably and precisely using a simple process model, even in the presence of large, global perturbations to the cell culture. In this work, we provide the implementation details of this scheme and demonstrate its performance on simulation case studies based on the actual experiments.