In this work, we use particle swarm optimization (PSO) to adjust the parameters of a membrane computing (MC) model of a synthetic autoinducer-2 (AI-2) signalling system in genetically engineered Escherichia coli bacteria. Bacteria release, receive and recognize signalling molecules in order to exchange information. These signalling molecules are responsible for coordinating gene expression at the population level in response to various stimuli such as size of the population, nutrient availability and other biochemical signals. This bacterial cell-to-cell communication is known as Quorum Sensing (QS). AI-2, from Vibrio harveyi, is the signaling molecule of interest in this study. We present a non-deterministic in silico model of Autoinducer-2 Quorum Sensing that is formalized by membrane computing (MC). The model is driven by 23 interaction rules that define biochemical interactions between independent compartments known as membranes. Due to the high dimensionality of this problem as well as lack of data relating to the biochemical parameters of this signalling system, we used a generic particle swarm optimization (PSO) algorithm to discover optimal solutions for the rule stochasticity constants. Our results were compared to the expected trends in quorum sensing behaviour. Ultimately, the results obtained from the PSO are thought to be in accordance with the predicted behaviour of the synthetic AI-2 signalling system.