Abstract
Background: Diffusion is a key component of many biological processes such as chemotaxis, developmental differentiation and tissue morphogenesis. Since recently, the spatial gradients caused by diffusion can be assessed in-vitro and in-vivo using microscopy based imaging techniques. The resulting time-series of two dimensional, high-resolutions images in combination with mechanistic models enable the quantitative analysis of the underlying mechanisms. However, such a model-based analysis is still challenging due to measurement noise and sparse observations, which result in uncertainties of the model parameters. Methods: We introduce a likelihood function for image-based measurements with log-normal distributed noise. Based upon this likelihood function we formulate the maximum likelihood estimation problem, which is solved using PDE-constrained optimization methods. To assess the uncertainty and practical identifiability of the parameters we introduce profile likelihoods for diffusion processes. Results and conclusion: As proof of concept, we model certain aspects of the guidance of dendritic cells towards lymphatic vessels, an example for haptotaxis. Using a realistic set of artificial measurement data, we estimate the five kinetic parameters of this model and compute profile likelihoods. Our novel approach for the estimation of model parameters from image data as well as the proposed identifiability analysis approach is widely applicable to diffusion processes. The profile likelihood based method provides more rigorous uncertainty bounds in contrast to local approximation methods.
Introduction
Biological patterning, such as the migration of immune cells to lymph nodes or embryonic development, is fundamentally driven by spatial gradients of signaling molecules. While modern microscopy provides high-resolution 2D and 3D visualizations of these gradients, translating these images into predictive mathematical models remains a significant challenge. Primary obstacles include high levels of measurement noise, sparse temporal observations, and the inherent complexity of parameter identifiability in partial differential equation (PDE) models.
Methodology
The proposed framework treats individual image pixels as data points within a Maximum Likelihood Estimation (MLE) problem. The core components of the approach are:
- Mechanistic Modeling: Reaction-diffusion PDEs describe the spatiotemporal dynamics of molecules, including movement, substrate binding, and degradation.
- Likelihood Formulation: A specialized likelihood function accounts for multiplicative log-normal noise, which better reflects the positive-only and signal-dependent nature of microscopy pixel intensities compared to standard Gaussian assumptions.
- PDE-Constrained Optimization: The "inverse problem" of parameter estimation is solved by ensuring that the estimated parameters remain physically consistent with the underlying diffusion laws.
- Identifiability Analysis: The study utilizes Profile Likelihoods (PL) to assess parameter certainty. By systematically varying one parameter and re-optimizing others, the method distinguishes between "practically identifiable" parameters (well-constrained by data) and "non-identifiable" ones.
Case Study: Haptotaxis of Dendritic Cells
To validate the framework, the authors modeled the guidance of dendritic cells (DCs) toward lymphatic vessels (LVs), a process known as haptotaxis. This process relies on the cytokine CCL21, which exists in both soluble and tissue-immobilized forms.
The study employed a realistic artificial dataset mimicking this setup, featuring a Y-shaped source term representing branched vessels. The model maps continuous PDE solutions to discrete pixels by integrating concentration over the area of each pixel, directly linking mathematical theory to camera sensor reality.
Results and Discussion
The framework was used to estimate five key kinetic parameters: the diffusion constant ($D$), release rate ($\alpha$), binding rate ($k_1$), unbinding rate ($k_{-1}$), and degradation rate ($\gamma$).
The analysis revealed that while four parameters were practically identifiable with tight confidence intervals, the degradation rate ($\gamma$) remained non-identifiable under the initial experimental conditions. The profile likelihood for $\gamma$ was flat, indicating that a wide range of values could explain the observed data. However, the authors demonstrated that increasing the number of temporal observations ($M$) eventually allows for the identification of even these challenging parameters.
Furthermore, the study showed that traditional "Hessian-based" local approximations often underestimate uncertainty in diffusion models. Profile likelihoods provide a more rigorous and "honest" assessment of parameter reliability.
Conclusion
This research provides a robust pipeline for extracting quantitative insights from biological imaging. By integrating PDE-constrained optimization with profile likelihood analysis, researchers can move beyond qualitative observations to rigorous, uncertainty-aware modeling. Future work will focus on implementing adaptive mesh schemes for complex geometries and extending the framework to 3D voxel-based image stacks. Crucially, this method can serve as a tool for experimental design, allowing researchers to optimize imaging intervals before entering the lab to ensure parameter identifiability.
Summary updated