Investigating the growth of an engineered strain of Cyanobacteria with an Agent-Based Model and a Recurrent Neural Network

Abstract

A computational framework combining Agent-Based Models (ABMs) and Deep Learning techniques was developed to help design microbial communities that convert light and CO2 into useful bioproducts. An ABM that accounts for CO2, light, sucrose export rate and cell-to-cell mechanical interactions was used to investigate the growth of an engineered sucrose-exporting strain of Synechococcus elongatus PCC 7942. The ABM simulations produced population curves and synthetic images of colony growth. The curves and the images were analyzed, and growth was correlated to nutrients availability and colonies' initial spatial distribution. To speed up the ABM simulations, a metamodel based on a Recurrent Neural Network, RNN, was trained on the synthetic images of growth. This metamodel successfully reproduced the population curves and the images of growth at a lower computational cost. The computational framework presented here paves the road towards designing microbial communities containing sucrose-exporting Synechococcus elongatus PCC 7942 by exploring the solution space in silico first.