A Multi-agent approach for $\textit{in silico}$ simulations of micro-biological systems

Abstract

Using a Multi-agent systems paradigm, the present project develops, validates and exploits a computational testbed that simulates micro-biological complex systems, namely the aggregation patterns of the social amoeba Dyctiostelium discoideum. We propose a new design and implementation for managing discrete simulations with autonomous agents on a microscopic scale, thus focusing on their social behavior and mutual interactions. Then, the dependence on the main physical variables is tested, namely density and number of amoebas; in addition, we analyze the robustness of the dynamics against various noise sources. Along with these results, we suggest a methodology for further studies that make use of our validated model.

Dynamical strategies for obstacle avoidance during Dictyostelium discoideum aggregation: a Multi-agent system model

“…Four main agents types compose the model architecture: Environment Env (a squared closed dim × dim domain, composed by cells with associated food sources b(t) possibly growing over time), Amoeba Am (proactive agents representing individual cells), cAMP (vectorial packages of chemical signal) and Obstacles Obs. It has been shown [48] that the vectorial message-sharing approach generates results that are consistent to those obtained with the typical diffusion of chemicals, as we can substitute a chemical gradient ∇C = g ẑ with one that is associated to the flux of probability of sensing a discrete signal: P amoeba ({I starving cAM P : E g → per}), that is, the probability of a (starving) moving amoeba processing the information carried by an absorbed cAMP message. At the same time, such strategy is cheaper in terms of computational cost than diffusive ones [49,50].…”

“…In order to perform simulations and analysis, we make use of an existing MAS-based computational framework that was purposefully designed to address D. discoideum behavior and that has already been tested and validated [48]. Recall that such model involves the generation of the desired dynamic after stating individual behavioral rules and parameters.…”

“…We define a certain range (in turn defining a "neighborhood area") under which two cells can be said to be neighbors. The reason for defining a "range" is that, as adhesion cells are not implemented in the model, cells are kept close by social interactions (message sharing) [48]. As a consequence, two neighbor cells can vary their distance in time by oscillating around each other.…”

“…where R is the agent radius, v A is the agent speed, ∆t is equal to 1 cycle; r is the mean distance from the centers of two overlapping cells, that can be determined by studying the chosen simulations. Agent parameters and exact values come from settings described in [48].…”

“…This way we can compare what happens when obstacles are placed locally and whether the obstacle-free subspace elicits the same aggregation pattern as when n obs = 0. Consequently, after having set the colony parameters according to those that were validated in paper [48], a point in SP = {n obs × d im × x0 ∈ Sub Env × nat obs } is chosen.…”