Agent-Based Rules Reproduce Fibroblast Wound Closure With High Fidelity
Fibroblast scratch assays remain a staple of wound-healing research because they capture two core cellular behaviors at once: migration into an empty gap and proliferation behind the advancing edge. The drawback is familiar to any laboratory working in this space. Traditional scratch experiments take time, require repeated imaging, and can be difficult to reproduce when geometry, cell density, or local matrix conditions shift from one run to the next.
This preprint describes an agent-based model built to mimic those dynamics in silico. The simulation represents fibroblasts and the extracellular matrix as particles and encodes local rules for migration, proliferation, contact inhibition, cell-cell mechanical interactions, and ECM behavior. In practical terms, each cell acts on its immediate surroundings rather than on a global command, which mirrors how fibroblasts actually behave in a wound bed: they sense neighbors, respond to crowding, and move into available space while the matrix itself influences the physical path available for closure.
The modeling approach matters because scratch assays are not simply a measure of cell speed. Wound closure emerges from a balance between motility, division, mechanical resistance, and the shifting geometry of the scratch. A framework that can capture those interacting forces without constant experimental reruns offers a different kind of laboratory tool—one that is faster to iterate and easier to standardize across conditions.
Calibration Against Six Imaging Time Series Produced Strong Agreement
Investigators validated the model with two tests. The first was a leave-one-out analysis in which parameters were calibrated to six imaging time series and then tested against the seventh. Across the tested wound geometries, the model achieved a mean normalized root mean squared error of 3.34%, and Pearson’s R exceeded 0.98.
Those values indicate exceptionally close agreement between simulated and observed wound closure trajectories. In a system as spatially dynamic as a scratch assay, correlation alone would not be enough; a model can track the general trend and still miss the geometry-specific details that matter in real experiments. The low error here suggests that local rules were sufficient to recapitulate the observed behavior across different wound shapes rather than only one idealized assay configuration.
A one-sentence result stands out: the model tracked distinct wound geometries with near-unity correlation.
That matters because geometry changes the entire mechanical problem. A narrow scratch exposes a different leading edge than a broad one, and irregular borders alter how fibroblasts distribute across the wound front. A useful model has to absorb those differences without becoming dependent on hand-tuned assumptions for each individual assay.
Early Data Were Enough to Forecast Three Additional Days of Closure
The second validation test was even more operationally relevant. Parameters were calibrated to the first 24 hours of data, and the remaining 66 hours were predicted. In that prediction window, the model achieved an NRMSE of 2.08%.
Forecasting wound closure from early measurements is the kind of result that gives computational models practical value. Early scratch assay imaging is often the point at which researchers decide whether to keep a condition, adjust the assay design, or repeat an experiment. If a model can infer the rest of the closure trajectory from the first day, it can reduce wasted imaging time and help distinguish between true biological differences and noise from assay preparation.
The biological logic behind the performance is straightforward. Fibroblast closure does not depend on a single dominant mechanism; it emerges from short-range interactions repeated across thousands of cells. Once the model learns the local rules governing movement, crowding, and matrix interaction, the later stages of closure become more predictable because the same rules continue to govern edge advancement and wound filling.
Why Local Mechanical Rules Matter in Wound Biology
The model’s structure reflects a central principle in tissue repair: cells are not isolated agents moving independently through empty space. Fibroblasts encounter resistance from neighboring cells, alter their trajectories when contact inhibition increases, and deform or respond to the extracellular matrix while migrating. By representing these processes at the level of local interactions, the simulation can reproduce collective behavior without requiring a high-level global controller.
That is especially relevant for wound-healing biology, where the edge of a scratch is rarely smooth and the microenvironment can vary across millimeters. Mechanical coupling between cells and matrix can create wave-like patterns of closure, uneven edge advance, and local congestion near the wound front. A particle-based agent model is well suited to capture those effects because it can represent spatial heterogeneity directly instead of averaging it away.
The result is not just a numerical fit to an image sequence. It is a mechanistic framework that translates observed wound behavior into a reusable set of rules. Researchers can test how changing migration, proliferation, or contact inhibition alters closure without repeating every experiment at the bench.
Practical Value for In Silico Fibroblast Experiments
The authors position the model as an efficient platform for experimentation and in silico study of fibroblast behavior. That claim is grounded in the validation results, but the broader promise lies in flexibility. Scratch assays can be expensive in time and labor, and repeated imaging across multiple geometries can become a bottleneck. A validated computational model can narrow the number of wet-lab experiments needed to explore a hypothesis, especially when the main question concerns relative changes in cell behavior rather than absolute assay output.
This also has implications for how wound-healing studies are designed. If a model can reliably predict closure trajectories from early images, investigators may be able to use it to compare assay geometries, identify parameter regimes that produce ambiguous readings, or test how strongly a given biological perturbation must shift migration or proliferation before the wound trajectory changes measurably.
The work remains a preprint and has not yet undergone peer review. Even so, the reported performance metrics are strong enough to make the model worth watching as a potential scaffold for future fibroblast studies. The key unresolved issue is not whether a local-rule framework can fit the available data; it clearly can. The more important next step is whether the same structure will hold when the system becomes more complex, with additional cell states, broader biological perturbations, or experimental noise beyond the controlled scratch assays used here.