Figure 1. Schematic of artificial intelligence system for controlling evolving cell populations.

A: E. coli population evolving on fitness landscapes under the strong selection, weak mutation evolutionary regime. At each time step, a reward signal r and a measure of system state s are sent to the replay memory structure. B: Replay memory array stores (s, a, r, s’) tuples where s’ is state s+1. These are then used to batch train the neural network. C: Deep Neural network estimates the value of each action given information about the environment’s state. The action with the largest estimated value is then applied to the evolving cell population.