Development of a Parameter Estimation Pipeline for Multi-Cellular Biological Processes

Abstract

Tissue dynamics are complex and involve interactions between different cell types and extracellular components on different spatial and temporal scales. Specific tissue properties are relevant for a broad range of processes, including tissue homeostasis, viral infection, and tumor development and treatment. The multi-scale and multi-cellular model has been proven to be a valuable tool to study these dynamics. However, these models remain challenging to use in practice. The lack of computational tools and widely adopted data standards that facilitate its simulation and calibration hinders its wider adaptation.<br/> The goal of this cumulative thesis is to advance the computational foundation needed for multi-scale and multi-cellular models to become more accessible, scalable, and reproducible, with a special focus on the parameter estimation aspect. <br/> First, we develop the FitMultiCell pipeline for simulating and parameterizing multi-scale and multi-cellular biological systems. The pipeline has been tested in different real biological applications to validate its robustness and scalability. It provides a user interface for easier adaptation with an extensive set of documentation. <br/> Second, we improve the wall-time of the approximate Bayesian computation—the sequential Monte Carlo (ABC-SMC) inference method in the pipeline by introducing a new scheduling strategy where we are able to achieve better wall-time by introducing a higher resource utilization. We evaluate the new strategy across multiple realistic application scenarios and show consistent performance improvements. By shortening runtimes and improving scalability, the approach enables the calibration of more complex models within practical time limits. <br/> Third, we introduce a standardized format, PEtab-MS, to encapsulate the parameter estimation problem of a multi-scale and multi-cellular nature. PEtab-MS provides a structured, reusable, and machine-readable way to define the model, experimental data, conditions, parameters, and objective function in a reusable and machine-readable form. It enables easier sharing and comparison of results across different studies. This will improve collaboration and advance the research of these models. <br/> Fourth, we apply these advances to investigate the cell migration process in a pillar-forest microenvironment by using a cellular Potts model (CPM) calibrated on experimental data. The results not only show the impact of spatial constraints on impacting the migration dynamics but also highlight the potential of data-driven modeling pipelines to derive mechanistic insights from complex systems. <br/> In conclusion, the advances presented in this thesis enable wider adaptation through the introduction of a general-purpose pipeline, a standardized calibration format, and a faster inference strategy. The practical value of these contributions is demonstrated across multiple biological case studies, both individually and as an integrated end-to-end workflow.