A Multi-layer Neural Network Model to predict the Quorum Sensing behavior of a heterogeneous community of bacteria.*

Bacteria communicate through the exchange of molecular signals in a process known as quorum sensing. This process is well understood for populations containing only one species of bacteria. The release of a specific signaling molecule leads to a high concentration of the signal. At a high concentration of signal, the bacteria activate the expression of multiple quorum sensing-regulated genes. In many natural contexts, many species coexist, and some of these species may produce different chemical variants of a particular signal molecule. This leads to crosstalk between species, as signals produced by one species may promote or inhibit the activation of quorum sensing in another species. There are even examples of a species participating in two orthogonal signaling pathways, although often both signaling pathways activate similar sets of genes. Here bacterial communication given the added complexity of multiple species exchanging multiple types of signals is modeled as a multilayer neural network. The network has one or two layers, depending on the number of orthogonal layers of signal exchange utilized within the bacterial community. Nodes in the network represent cells belonging to a particular species with a weighted directed edge representing signal crosstalk. We investigate activity patterns of the nodes at steady state, representing decision states of the network. The robustness of these decision states is probed via perturbations of signal concentrations and the population size. Nodes that are more influential in setting the community decision state are identified. For a network with two layers of signal, the correlation of activity states in each layer is explored and information exchange between the two layers of the network is quantified. The presence of crosstalk and multiple layers of orthogonal has consequences in the overall percentage of active nodes as well as the stability of these networks.