Biochemical Oscillations in HeLa Cells: Automated Time-Lapse Image Analysis and Two-Step Parameter Estimation

In recent times, single-cell analysis from time-lapse imaging has been used to characterize distinct subpopulations of a complex heterogeneous cell system. This can offer significant insights leading to robust cancer treatment strategies. In particular, calcium imaging may have significance since the pattern in oscillation may encode the cell state. Although model fitting to single-cell dynamics is pivotal, the analysis remains time consuming due to manual segmentation from time-lapse microscopic images. Secondly, the parameter estimation for the nonlinear system of differential equations (ODEs) describing these oscillations remains challenging due to the massive search space and the presence of multiple local minima in the objective function. To address these problems,firstly, we propose a computer vision algorithm (YOLOv9) to automatically extract calcium signals from the live cell imaging video dataset. Next, we implemented a two-step approach that combines Monte Carlo simulations (MC) and Genetic algorithm (GA) for parameter estimation. The first step (MC), was used to narrow down the search space and find the best performing upper and lower bounds for GA. In the second step, we implemented GA to determine the optimized parameters for the nonlinear model. Monte Carlo sampling along with GA shows significant reduction in KL divergence compared to only GA. We have also presented multiple simulations for the stochastic version of the model using the estimated parameters. The proposed pipeline shows significant promise for understanding the underlying cell states.