Deep reinforcement learning for the control of bacterial populations in bioreactors

Multi-species bacterial communities are widespread in natural ecosystems. Engineered synthetic communities have shown increased productivity over single strains and allow for the reduction of metabolic load by compartmentalising bioprocesses between multiple sub-populations. Despite these benefits, co-cultures are rarely used in practice because control over the constituent species of an assembled community has proven challenging. Here we demonstrate the efficacy of approaches from artificial intelligence -- reinforcement learning -- in the control of co-cultures within continuous bioreactors. We first develop a mathematical model of bacterial communities within a chemostat that incorporates generalised Lotka-Volterra interactions. We then show that reinforcement learning agents can learn to maintain multiple species of cells in a variety of chemostat systems, with different dynamical properties, subject to competition for nutrients and other competitive interactions. Reinforcement learning was also shown to have the ability to maintain populations within more selective bounds, which is important for optimising the productivity of reactions taking place in co-cultures. Additionally, our approach was shown to generalise to systems of three populations. These systems represent a level of complexity that has not yet been tackled by more traditional control theory approaches. As advances in synthetic biology increase the complexity of the cellular systems we can build, the control of complex co-cultures will become ever more important. Data-driven approaches such as reinforcement learning will enable greater optimisation of environments for synthetic biology.