Probabilistic reconstruction of truncated particle trajectories on a closed surface

Yunjiao Lu, Pierre Hodara, Charles Kervrann, Alain Trubuil

Abstract

Investigation of dynamic processes in cell biology very often relies on the observation in two dimensions of 3D biological processes. Consequently, the data are partial and statistical methods and models are required to recover the parameters describing the dynamical processes. In the case of molecules moving over the 3D surface, such as proteins on walls of bacteria cell, a large portion of the 3D surface is not observed in 2D-time microscopy. It follows that biomolecules may disappear for a period of time in a region of interest, and then reappear later. Assuming Brownian motion with drift, we address the mathematical problem of the reconstruction of biomolecules trajectories on a cylindrical surface. A subregion of the cylinder is typically recorded during the observation period, and biomolecules may appear or disappear in any place of the 3D surface. The performance of the method is demonstrated on simulated particle trajectories that mimic MreB protein dynamics observed in 2D time-lapse fluorescence microscopy in rod-shaped bacteria.

Introduction

In cell biology, 2D microscopy is frequently used to study 3D processes, such as the movement of MreB proteins along the inner membrane of rod-shaped bacteria. Because these bacteria are roughly cylindrical, standard 2D microscopy techniques like TIRFM (Total Internal Reflection Fluorescence Microscopy) only capture the portion of the membrane closest to the glass slide.

Conceptual model of 2D microscopy capturing 3D cylindrical trajectories, showing the truncation of paths as they move to the 'dark side' of the cell.

When a protein moves toward the unobserved "back" of the cell, it vanishes from the field of view. It may later reappear on the other side or dissociate ("die") while out of sight. Traditional tracking algorithms struggle with these "re-entrance" events, leading to fragmented trajectories. This paper introduces a probabilistic framework to "stitch together" these broken segments by combining stochastic motion modeling with Integer Linear Programming (ILP).

Methodology

The proposed reconstruction method follows a structured pipeline to move from raw observed "tracklets" to fully reconstructed global trajectories.

Flowchart of the probabilistic reconstruction framework, detailing the steps from parameter estimation to global optimization.

1. Probabilistic Modeling

Particle movement is modeled as Brownian motion with drift (constant velocity plus random diffusion). The "birth" and "death" of particles are modeled using Poisson processes. The core innovation is a likelihood function $Q(c)$ that calculates the probability of a specific configuration—a set of matches between particles exiting the visible region ("outputs") and those entering it ("inputs"). The model explicitly accounts for three possibilities: a particle died in the dark zone, it is a brand-new arrival, or it is a specific particle that previously exited.

2. Global Optimization

To find the most likely configuration, the problem is formulated as an Integer Linear Programming (ILP) task. This allows the researchers to solve for the global maximum likelihood efficiently using standard solvers like CPLEX, ensuring that the final reconstruction is the most statistically probable set of connections.

3. Parameter Estimation

The method is data-driven and includes novel estimators to automatically calculate drift, variance, and birth/death rates directly from the observed tracklets. This removes the need for manual parameter tuning, making the tool more accessible for experimental biologists.

Key Results and Biological Application

The performance of the algorithm was validated using both synthetic and real biological datasets:

  • Synthetic Validation: Using simulated trajectories on an unwrapped cylindrical surface, the method achieved Adjusted Rand Index (ARI) scores > 0.9 for moderate particle densities. Even at high densities (up to 100 tracklets per movie), the median ARI remained above 0.7, significantly outperforming simple heuristic assignments.
  • Real Biological Data: The method was applied to TIRFM movies of MreB aggregates in Bacillus subtilis. The algorithm successfully reconstructed trajectories showing proteins making multiple full rotations around the bacterial cell, providing a more complete picture of the dynamics involved in cell wall construction.

Conclusion

By treating the cylindrical topology as a periodic boundary and applying a rigorous probabilistic framework, this research enables the recovery of full 3D dynamics from partial 2D observations. While the current model assumes constant velocity and spatial homogeneity, it provides a robust foundation for future extensions involving non-homogeneous surfaces and more complex, age-dependent particle lifetimes.

Summary updated

Abstract

Investigation of dynamic processes in cell biology very often relies on the observation in two dimensions of 3D biological processes. Consequently, the data are partial and statistical methods and models are required to recover the parameters describing the dynamical processes. In the case of molecules moving over the 3D surface, such as proteins on walls of bacteria cell, a large portion of the 3D surface is not observed in 2D-time microscopy. It follows that biomolecules may disappear for a period of time in a region of interest, and then reappear later. Assuming Brownian motion with drift, we address the mathematical problem of the reconstruction of biomolecules trajectories on a cylindrical surface. A subregion of the cylinder is typically recorded during the observation period, and biomolecules may appear or disappear in any place of the 3D surface. The performance of the method is demonstrated on simulated particle trajectories that mimic MreB protein dynamics observed in 2D time-lapse fluorescence microscopy in rod-shaped bacteria.
BibTeX
@article{lu2021probabilistic,
  author = {Yunjiao Lu and Pierre Hodara and Charles Kervrann and Alain Trubuil},
  title = {{Probabilistic reconstruction of truncated particle trajectories on a closed surface}},
  journal = {Multiscale Model. Simul.},
  year = {2021},
  doi = {10.1137/20m1333742},
}
Figure