🔬 condensed matter

A framework for modeling and inferring tracer diffusion in crowded environments

This paper presents a framework combining minimal simulations and a parallel partial Gaussian process model to rapidly infer structural properties and predict tracer diffusion in crowded environments, revealing that transport is governed by accessible pore sizes and can be applied to both soft matter suspensions and living cells.

Original authors: Jinseok Lee, Tong Lin, Mengyang Gu, Yimin Luo

Published 2026-05-07

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Jinseok Lee, Tong Lin, Mengyang Gu, Yimin Luo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to walk through a crowded room. If the room is empty, you can stroll freely in any direction. But as you fill the room with people, your path becomes harder. You have to weave around them, and eventually, you might get stuck in a small pocket of space, unable to move far from where you started.

This paper is about understanding exactly how that "crowded room" affects the movement of tiny particles (called tracers) in two very different settings: a lab-made mixture of soft gels and the inside of a living cell.

Here is the story of their discovery, broken down into simple parts:

1. The Problem: The "Crowded Room" Mystery

Scientists have long known that things move differently in crowded places. In biology, the inside of a cell is packed with organelles, proteins, and structures, making it a very crowded environment.

The Old Way: To study this, scientists usually track individual particles with a microscope and draw lines connecting their dots. But this is like trying to follow a single person in a mosh pit; it's hard to know who is who, and the software often gets confused.
The New Way: The authors developed a better way to look at the data without getting bogged down in tracking every single dot. They used a statistical method (called AIUQ) that looks at the "blur" of movement in the whole image to figure out how fast things are moving, without needing to name every particle.

2. The Experiment: Building a Crowded Room

The team created two types of "crowded rooms" to test their ideas:

The Gel Room: They mixed tiny, hard beads (the tracers) into a soup of soft, squishy hydrogel balls. By adding more and more soft balls, they made the room more crowded.
The Cell Room: They put tiny fluorescent beads inside living mouse cells. These beads got swallowed by the cell and had to navigate the clutter of the cell's interior.

What they found:
As they added more "furniture" (the soft balls or cell parts), the tracers stopped moving in straight lines. They went from free running to jittering in place, and finally to being trapped in small cages. There was a specific "tipping point" where the room became so full that the tracers could no longer travel freely.

3. The Simulation: The "Digital Twin"

To understand why this happened, they built a computer simulation.

The Mistake: At first, they just told the computer, "Put these balls here, and let the small bead bounce around." But the computer's results didn't match the real experiment. The bead in the computer moved too fast.
The Fix: They realized they forgot about friction. In the real world, when a particle gets close to a wall or another particle, the fluid between them gets squished, creating extra drag (like trying to slide your hand through water right next to a glass wall).
The Result: Once they added this "hydrodynamic drag" to the simulation, the computer model perfectly matched the real-life experiments. It showed that the bead wasn't just blocked by walls; it was also slowed down by the fluid squeezing between the obstacles.

4. The "Crystal Ball": The PPGP Model

Running the computer simulation was accurate, but it was slow. It took about 10 seconds to simulate one scenario. If they wanted to test thousands of different room layouts (different sizes of furniture, different crowd densities), it would take forever.

So, they trained a Machine Learning "Crystal Ball" (called a Parallel Partial Gaussian Process, or PPGP).

How it works: They fed the model 160 different simulation results. The model learned the pattern: "If the room is 50% full and the furniture is big, the particle moves this way."
The Superpower: Once trained, this model could predict how a particle would move in any new crowded room in less than a blink of an eye (25 milliseconds). It is about 100,000 times faster than running the full simulation.

5. The Big Surprise: Different Rooms, Same Movement

Here is the most fascinating part. The team tried to work backward: "If we see a particle moving this way, what does the room look like?"

They expected to find one specific answer. Instead, they found that many different room layouts look the same to the particle.

The Analogy: Imagine you are blindfolded in a maze. You might feel a wall on your left and a wall on your right. You could be in a narrow hallway, or you could be in a small square room. Your immediate feeling (the movement) is the same, even though the global shape of the maze is totally different.
The Conclusion: The particle only "feels" the size of the immediate hole it is standing in (the pore size). Two completely different crowds can have the same size of holes, so the particle moves the same way in both. This means you can't always tell exactly what the whole room looks like just by watching one particle, but you can tell how big the holes are.

6. Applying it to Cells

Finally, they tested if this "minimal model" worked for living cells.

They measured the size of the "furniture" inside the cell (organelles).
They used their fast model to predict how a 100-nanometer bead should move.
The Match: The prediction was spot on.
The Test: They then tried it with a larger bead (200 nanometers) without changing the model, just by telling the model the bead was bigger. The model still predicted the movement correctly.

Summary

The paper provides a new, fast, and accurate toolkit for understanding how things move in crowded places.

Crowding matters: As a space gets fuller, movement changes from free-flowing to trapped.
Fluid friction matters: You can't just count the obstacles; you have to account for the fluid squeezing between them.
Speed matters: Their new AI model predicts movement instantly, saving hours of computer time.
Local vs. Global: A particle only cares about the size of the immediate hole it's in, not the entire shape of the room. Different rooms can feel the same to a traveler.

This framework helps scientists understand how drugs or nutrients might move through the crowded interior of a cell, using the particle's movement as a map to understand the cell's internal structure.

Technical Summary: A Framework for Modeling and Inferring Tracer Diffusion in Crowded Environments

Problem Statement
Tracer diffusion in crowded environments, such as intracellular spaces and soft matter suspensions, is central to biological and physical processes. While tracer trajectories provide spatial and statistical information, quantitative frameworks linking tracer motion to environmental structure remain limited. Existing methods face significant challenges:

Multiple Particle Tracking (MPT): While widely used, MPT relies on user-defined parameters (e.g., search radius) that struggle in highly heterogeneous environments where spatial variability makes constant cutoffs inappropriate.
Fourier-based techniques (e.g., DDM): These avoid trajectory linking but often require specific wave vector ranges or model assumptions.
Inverse Problem: Determining environmental properties (e.g., area fraction, particle size, polydispersity) from tracer trajectories is difficult because individual trajectories are dominated by local geometry, and systematic variations in simulation parameters have led to divergent conclusions regarding which variables dominate diffusion regimes.
Computational Cost: Simulating tracer dynamics with sufficient physical fidelity to match experiments is computationally expensive, making broad parameter space exploration impractical.

Methodology
The authors propose a multi-stage framework combining experiments, minimal physics-based simulations, and data-driven predictive modeling:

Experimental Characterization:
- Soft Particle System: Fluorescent rigid tracers ( $2a = 2 \, \mu\text{m}$ ) were dispersed in suspensions of soft PEGDA hydrogel particles (average radius $\bar{R} \approx 4.35 \, \mu\text{m}$ ) at varying area fractions ( $\Phi$ ).
- Cellular System: Tracers ( $2a = 100, 200 \, \text{nm}$ ) were internalized into NIH3T3 cells to study transport in the crowded cytoplasm.
- Analysis: Trajectories were analyzed using both MPT and a model-free Ab Initio Uncertainty Quantification (AIUQ) framework. AIUQ was preferred for model calibration as it avoids parameter tuning, provides stable mean-squared displacement (MSD) curves over broad lag times, and handles intermittent hopping behavior better than MPT.
Minimal Simulation Framework:
- A Monte Carlo simulation was developed with $N_t$ tracer particles moving among $N_m$ polydisperse matrix particles in a 2D box.
- Physical Ingredients: The model incorporates steric exclusion (hard-sphere interactions) and hydrodynamic hindrance.
- Hydrodynamic Corrections: Crucially, the simulation includes position-dependent corrections to the diffusion coefficient ( $D$ $D$ ) to account for:
  - Wall hindrance (proximity to the glass substrate).
  - Lubrication-like drag near matrix particle surfaces (modeled as a gap-dependent correction: $D_{\parallel}/D_0 \sim (g^*/(g^*+g_0))^{\alpha_{hyd}}$ ).
- Without these hydrodynamic corrections, simulations failed to reproduce experimental MSDs, particularly at small lag times and intermediate area fractions.
Data-Driven Predictive Modeling (PPGP):
- A Parallel Partial Gaussian Process (PPGP) model was trained on simulation outputs.
- Inputs: Matrix area fraction ( $\Phi$ ), average radius ( $\bar{R}$ ), and polydispersity ( $p$ ).
- Output: The vector of MSD curves (represented by the logarithmic slope $\alpha$ ).
- Performance: The PPGP model predicts MSDs in $\sim 25$ ms, accelerating predictions by $\sim 10^5$ times compared to running full simulations or experiments.
Inverse Inference and Sensitivity Analysis:
- The framework was used to map observed MSDs back to structural variables.
- Global sensitivity analysis (Sobol's indices) and regression were performed to determine the influence of $\Phi$ , $\bar{R}$ , and $p$ on MSD accuracy.

Key Results

Diffusive-to-Confined Transition: Experiments and simulations confirmed a transition from diffusive to subdiffusive (confined) motion as the matrix area fraction $\Phi$ increases. A distinct plateau in MSD emerges at high $\Phi$ ( $\approx 0.5$ ), indicating restricted accessible pore space.
Necessity of Hydrodynamics: Simulations only quantitatively matched experimental MSDs and Van Hove correlation functions when hydrodynamic corrections (wall and particle proximity effects) were included. Steric exclusion alone was insufficient.
Percolation Threshold ( $\Phi_c$ ): The PPGP model enabled rapid identification of a tracer-defined percolation threshold $\Phi_c$ $Φ_{c}$ (the "knee" where confinement significantly reduces long-time diffusion).
- $\Phi_c$ increases with both average matrix radius ( $\bar{R}$ ) and polydispersity ( $p$ ).
- The tracer-defined threshold occurs at lower area fractions than the geometric percolation of void space because tracers are excluded from narrow necks between matrix particles.
Non-Uniqueness of Inverse Solution: The analysis revealed that distinct global matrix structures can produce indistinguishable MSDs. Specifically, different combinations of $\Phi$ $Φ$ , $\bar{R}$ $\overset{ˉ}{R}$ , and $p$ $p$ can yield nearly identical pore size distributions, leading to similar tracer dynamics.
- Sensitivity: The area fraction $\Phi$ is the dominant factor driving MSD variability ( $|\beta_\Phi| \approx 0.87$ ), followed by $\bar{R}$ . Polydispersity $p$ plays a minor role in isolation but interacts significantly with $\bar{R}$ .
- Implication: If $\bar{R}$ is known, $\Phi$ can be inferred with greater accuracy; otherwise, the inverse problem has multiple solutions corresponding to similar local pore geometries.
Application to Intracellular Transport: The minimal model successfully captured the MSDs of tracers in living cells. By assuming a cytoplasmic viscosity $\sim 100\times$ $\sim 100 \times$ water and a matrix mobility factor $m \approx 0.3$ $m \approx 0.3$ , the model reproduced experimental data for $100$ nm tracers.
- Extrapolation: The model, calibrated on $100$ nm tracers, accurately predicted the MSDs of $200$ nm tracers without re-parameterization, validating the model's ability to capture scale-dependent confinement effects.

Significance and Claims
The paper claims to establish a quantitative framework that links tracer dynamics to the accessible geometry of crowded environments across different scales.

Efficiency: The PPGP model allows for rapid screening of the parameter space, reducing prediction time from hours (experiment) or seconds (simulation) to milliseconds.
Physical Insight: The work demonstrates that tracer transport is primarily governed by accessible pore sizes rather than global structural details. This explains why different global configurations can yield identical MSDs.
Generalizability: The minimal model, incorporating steric and hydrodynamic effects, is sufficient to describe transport in both passive soft particle suspensions and active intracellular environments, provided matrix mobility is accounted for.
Inverse Problem: The study highlights the limitations of inferring global structural properties solely from tracer MSDs due to the non-uniqueness of the solution, suggesting that combining MSD analysis with independent measurements (e.g., of particle size) is necessary for accurate inference.

The authors conclude that this framework enables better formulation and understanding of transport in crowded environments, with specific utility in characterizing intracellular transport where direct measurement of matrix area fraction is difficult.