Almet et al. of cells to being pushed). We call this motility rule smart shoving. We examine whether agentCbased simulations of different shoving mechanisms can be distinguished on the basis of single realisations and averages over many realisations. We emphasise the difficulty in distinguishing cell mechanisms from cellular automata simulations based on snapCshots of cell distributions, siteCoccupancy averages and the evolution of the number of cells of each species averaged over many realisations. This difficulty suggests the need for higher resolution cell tracking. Introduction Cellular migration in living tissue necessarily involves the motile cell interacting with other cells that compete with it for space and potentially impede its motion. Successful Proglumide sodium salt migration requires the displacement of other cells and may require remodelling of extracellular matrix. Fully detailed modelling of such processes requires attention to chemical and mechanical signals between the motile OCTS3 cell and its environment and the shapes of the motile cell and its neighbours. In contrast, simpler models are capable of providing insights into these subtle and complex problems. AgentCbased models are especially useful, as they enable various model effects to be incorporated in a relatively simple way, facilitating experiments related to morphogenesis and colonisation in embryonic development [1, 2], wound healing [3], and tumour growth and metastasis in cancer [4C7]. An example of the utility of agentCbased modelling to the understanding of diseases is summarised in Landman et al. [8] where the incomplete invasion of the embryonic gastrointestinal mesenchyme by neural crest cells deprives the distal intestine of neurons, a condition called Hirschsprungs disease. A mathematical model of cell invasion, where motile cells also proliferate, successfully predicted invasion outcomes to imagined manipulations that were later verified experimentally. It is important to emphasise that Proglumide sodium salt the complexity of biological processes demands that careful attention is paid to model selection before attempting to simulate biological processes computationally. It particular, the model chosen must be capable of capturing the essence of the process being studied. It is also important to know whether there is any redundancy. Knowing which features of the model may be discarded and still yield satisfactory concordance with experimental observations gives important information not only on the model chosen, but also on the biological process and the sensitivity of the experimental measurements to capture the process of interest. In this study we will examine the ease with which different agentCbased motility mechanisms can be distinguished using metrics closely related to biological measurements. A motivating example for our approach is the experimental work reported by Iwanicki et al. [9] and Davidowitz et al. [10]. They studied an invasion process in which small clusters of ovarian cancer cells placed on top of an epithelial cell monolayer (grown on a suitable tissue culture substrate) force their way into the epithelial cell layer. This is a simple example of a more general problem in which a relatively thin layer of tissue is invaded by motile cells. We do not purport to model the ovarian cancer cell experiments specifically here, but rather to investigate more broadly model selection and redundancy for invasion problems. If we were concerned with detailed modelling of invasion into tightly constrained tissue, for which cells undergo large deformation and squeeze through interstices rather than moving into vacant space or simply displacing other agents, or use of structureless agents to represent cells would be an excessively crude approximation. Although invasion processes can be modelled using deterministic equations in which space and time are continuous, such approaches cannot shed light on the extent of variability in outcomes in the presence of the very real spatial and temporal stochasticity of motile biological cell populations. In contrast, each experiment on an agent-based model shows the locations of all cells in the model system. Averaging over large numbers of experiments with agent-based models gives access to similar information to that which one can obtain by deterministic continuum modelling (see the Appendix). There have been many recent papers on agentCbased models with potential application to development or invasion processes implemented on regular lattices. Typically, such models involve randomly moving agents (representing cells) subject to an exclusion process [11] in which attempted agent moves that would place an agent on an Proglumide sodium salt already occupied site are aborted. The probabilities of selection of which moves are to be attempted can also be allowed to depend in some way on the occupancy status.