Using artificial intelligence to identify spatio-temporal mechanisms of cell competition

Embryonic stem cells, also called pluripotent stem cells, are a cell type that exists in the early mammalian embryo and that have the potential to give rise to all the cell types and tissues that will form the foetus. The process by which these cells start to form more specialised cells is called differentiation and is very error prone. For this reason there are strict quality controls in place that prevent the emergence of abnormal embryonic stem cells. One such control is cells competition, that acts by comparing the fitness of cells within the stem cell population and eliminating those cells that are less fit than their neighbours. The elimination of abnormal cells by cell competition ensures that only the fittest cells go on to form the later embryo. Importantly, cell competition has also been shown to regulate cell fitness a variety of other contexts, from cancer to ageing. However, despite the importance of cell competition, we still do not understand the parameters that determine cell fitness, and therefore that establish which cells are the winners and which are losers of this fitness competition. The main difficulty lies in the feedback between cell rearrangements in the tissue and cell competition elimination; requiring to identify the role that cell proliferation, cell motility, apoptosis and cell signalling have during the competition process. To tackle this problem, we will develop a computational model of cell competition that will use the power of artificial intelligence to identify the contribution of different mechanisms of competition from live cell imaging in an unbiassed way. The artificial intelligence inference will identify the parameters that then we will verify experimentally to establish the most important factors that determine the outcome of cell competition. The identification of these factors will provide for the first-time insight into the rules that govern the competitive behaviour of pluripotent stem cells, and future work will be aimed at studying what other cell types abide by these rules, and therefore how universal these rules are.

Fitness control through cell competition is conserved from Drosophila to mammals and is present in a broad range of contexts, from growth control in the developing embryo to tissue homeostasis in the adult. Furthermore, it's deregulation has been implicated in cancer and organismal ageing. We have shown that during the onset of pluripotent stem cell differentiation cell competition eliminates around 35% of cells. Despite this importance, what parameters determine if a cell will be a winner and survive, or a loser and be eliminated during cell competition is poorly understood. One of the main challenges is that cell fitness is an emergent property in which cell-cell interaction, cell rearrangements, proliferation and apoptosis feedback into each other; impeding a straightforward analysis of the contribution of different mechanisms for cell competition. To tackle this challenge, we will generate an agent-based model of cell competition, with cell behaviour parameters that will be inferred using neural network assisted approximate Bayesian computation. Using this inference tool with live cell imaging data we will perform an unbiased search for potential parameters that can reproduce the outcomes of cell competition that we observe during the onset of pluripotent stem cell differentiation. This will include data from winner and loser pluripotent stem cells during normal competition, as well as in scenarios where the winners cannot eliminate the losers to establish which parameters are the most important to determine the outcome of cell competition. The predictions of the model will be tested and explored experimentally resulting in an experiment/computation/prediction cycle that will allow to identify robustly the mechanisms of cell competition. This work will help us uncover the rules that govern the competitive interactions occurring during early development, as well as providing us with tools to probe if these rules are universally conserved across tissues.