Cell-mechanical parameter estimation from 1D cell trajectories using simulation-based inference

Trajectories of motile cells represent a rich source of data that provide insights into the mechanisms of cell migration via mathematical modeling and statistical analysis. However, mechanistic models require cell type dependent parameter estimation, which in case of computational simulation is technically challenging due to the nonlinear and inherently stochastic nature of the models. Here, we employ simulation-based inference (SBI) to estimate cell specific model parameters from cell trajectories based on Bayesian inference. Using automated time-lapse image acquisition and image recognition large sets of 1D single cell trajectories are recorded from cells migrating on microfabricated lanes. A deep neural density estimator is trained via simulated trajectories generated from a previously published mechanical model of cell migration. The trained neural network in turn is used to infer the probability distribution of a limited number of model parameters that correspond to the experimental trajectories. Our results demonstrate the efficacy of SBI in discerning properties specific to non-cancerous breast epithelial cell line MCF-10A and cancerous breast epithelial cell line MDA-MB-231. Moreover, SBI is capable of unveiling the impact of inhibitors Latrunculin A and Y-27632 on the relevant elements in the model without prior knowledge of the effect of inhibitors. The proposed approach of SBI based data analysis combined with a standardized migration platform opens new avenues for the installation of cell motility libraries, including cytoskeleton drug efficacies, and may play a role in the evaluation of refined models.