Predicting Epithelial-Mesenchymal Transition Dynamics Under Specific Perturbations Using Data Assimilation

Epithelial-mesenchymal transition (EMT) is the fundamental transdifferentiation process of an epithelial cell to a mesenchymal cell, which includes losing epithelial-type cell-cell adhesion and gaining the mesenchymal-type enhanced cell motility. EMT has been observed to be a multistep process with partial EMT states, for which the cell expresses both epithelial and mesenchymal markers simultaneously. These partial EMT states play a central role in pathological conditions such as cancer metastasis. Experimentally, computational models have served as a valuable tool to predict mechanisms governing EMT phenotype transitions. However, long-term model predictions for an individual experiment often fail due to parameter uncertainty, since the parameters are determined using averages or extrapolations from multiple experimental conditions. Thus, model simulations generally represent typical responses of the system to perturbations, not necessarily the specific experiment responses of an individual system. Here, we propose a data-assimilation approach that improves long-term predictions by combining limited noisy observations with predictions from a computational model of TGF-β-induced EMT. We tested our approach in a series of “synthetic” in silico experiments, in which a forecasting system was tasked with reconstructing EMT dynamics of a phenotypically different “true” system and predicting the response to a series of perturbations, specifically time-varying inputs of exogenous TGF-β doses. We found that the forecasting system - with optimal data-assimilation algorithmic parameters - successfully predicted the response of the true system and accurately reconstructed its final phenotypical cell state (either epithelial, mesenchymal, or partial EMT). Future work will include testing and optimizing our approach over a wider range of physiological perturbations that will promote or suppress EMT and ultimately testing against in vitro experimental data.