Traction force microscopy with optimized regularization and automated Bayesian parameter selection for comparing cells

Yunfei Huang, Christoph Schell, Tobias B. Huber, Ahmet Nihat Şimşek, Nils Hersch, Rudolf Merkel, Gerhard Gompper, Benedikt Sabass

TL;DR

This paper tackles the "subjectivity gap" in Traction Force Microscopy (TFM) by introducing automated Bayesian methods for parameter selection. By moving away from manual, "eye-balled" regularization, the authors provide a robust framework for objectively comparing cellular forces across different experimental conditions and time-series data.

What Problem Does This Paper Address?

Cells are mechanical engines; they pull and push on their environment to move, divide, and maintain tissue structure. To measure these tiny forces, researchers use Traction Force Microscopy (TFM), where cells are placed on a soft, elastic gel filled with fluorescent beads. By tracking how the beads move (displacement), we can work backward to calculate the forces the cell exerted (traction).

Mathematically, this is an "inverse problem"—and a notoriously "ill-conditioned" one. Small amounts of experimental noise in the bead tracking can lead to massive, nonsensical errors in the calculated force field. To fix this, researchers use regularization, a mathematical technique that penalizes "noisy" or "unphysical" solutions.

The catch? Regularization requires a "tuning parameter" ($\lambda$). Currently, most researchers pick this parameter manually or use heuristics like the "L-curve" method, which often fails in real-world noisy data. This makes TFM results subjective: two different researchers might "tune" the same data to show different force magnitudes, making it nearly impossible to reliably compare cells in different conditions.

Methods

The authors systematically evaluated several reconstruction strategies, categorized into two main groups:

  1. Advanced Regularization Schemes: They tested classical L1 (Lasso) and L2 (Ridge) methods against three techniques borrowed from computer vision:
  2. Elastic Net (EN): A hybrid that combines the strengths of L1 and L2.
  3. Proximal Gradient Lasso (PGL) & Elastic Net (PGEN): Wavelet-based approaches designed to smooth out high-frequency noise.
  4. Bayesian Inference Models: They developed two new automated methods:
  5. Bayesian L2 (BL2): Automatically determines the optimal regularization by treating the parameter as a random variable to be inferred, requiring only an independent estimate of measurement noise.
  6. Advanced Bayesian L2 (ABL2): A fully "hands-off" version that infers both the optimal regularization and the noise level directly from the data.

Datasets

The study utilized a mix of synthetic and biological data:

  • Artificial Test Data: Simulated traction fields with 9 to 15 circular "force spots" of varying magnitudes and added Gaussian noise. This allowed the authors to know the "ground truth" and precisely measure error.
  • Primary Murine Podocytes: Kidney cells studied on polyacrylamide gels (16 kPa).
  • Rat Embryonic Heart Muscle Cells: Cardiomyocytes on silicone elastomers (15 kPa), used specifically to test time-series analysis during cell contraction.

Key Insights and Ideas

  • The "Fishing Net" Advantage: The authors found that Elastic Net (EN) is the most accurate method for static images. It acts like a "stretchable fishing net," retaining significant force signals while effectively filtering out small background noise.
  • Simpler is Better in Bayesian TFM: Interestingly, the authors tested complex hierarchical Bayesian models (often used in machine learning) but found they performed poorly for TFM, often overfitting to noise. The simpler global prior used in BL2/ABL2 proved much more robust.
  • Solving the Time-Series Dilemma: In a beating heart cell, force levels change rapidly. Using a single fixed regularization parameter for the whole movie sacrifices accuracy. The Bayesian approach (BL2) automatically adapts the parameter for every single frame, ensuring the reconstruction remains faithful even as the signal-to-noise ratio fluctuates.

Key Results

  • Accuracy: Elastic Net (EN) outperformed all other methods in reconstruction accuracy, particularly for undersampled data.
  • Automation vs. Subjectivity: BL2 and ABL2 produced results comparable to the "best possible" manually tuned L2 regularization but removed the human element entirely.
  • Computational Trade-offs: There is no free lunch. While L2 regularization takes only ~8 seconds, the superior Elastic Net (EN) requires roughly 48 minutes (0.8 h) and significantly more RAM (3.87 GB vs 350 MB) for the same dataset.
  • Robustness: Bayesian methods remained reliable even when noise levels were high ($\sigma_n/\sigma_u \approx 1$), a regime where traditional L-curve selection typically fails.

Limitations and Future Work

The primary limitation is computational cost. The most accurate method (Elastic Net) and the most automated method (ABL2) are significantly slower than standard L2 methods. While manageable for single cells, this could become a bottleneck for high-throughput screening.

The authors suggest that future work could involve applying these Bayesian frameworks to 3D traction force microscopy or more complex substrate geometries, where the "subjectivity problem" of manual tuning is even more pronounced.

Summary updated

Abstract

Adherent cells exert traction forces on to their environment which allows them to migrate, to maintain tissue integrity, and to form complex multicellular structures during developmental morphogenesis. Traction force microscopy (TFM) enables the measurement of traction forces on an elastic substrate and thereby provides quantitative information on cellular mechanics in a perturbation-free fashion. In TFM, traction is usually calculated via the solution of a linear system, which is complicated by undersampled input data, acquisition noise, and large condition numbers for some methods. Therefore, standard TFM algorithms either employ data filtering or regularization. However, these approaches require a manual selection of filter- or regularization parameters and consequently exhibit a substantial degree of subjectiveness. This shortcoming is particularly serious when cells in different conditions are to be compared because optimal noise suppression needs to be adapted for every situation, which invariably results in systematic errors. Here, we systematically test the performance of new methods from computer vision and Bayesian inference for solving the inverse problem in TFM. We compare two classical schemes, L1- and L2-regularization, with three previously untested schemes, namely Elastic Net regularization, Proximal Gradient Lasso, and Proximal Gradient Elastic Net. Overall, we find that Elastic Net regularization, which combines L1 and L2 regularization, outperforms all other methods with regard to accuracy of traction reconstruction. Next, we develop two methods, Bayesian L2 regularization and Advanced Bayesian L2 regularization, for automatic, optimal L2 regularization. Using artificial data and experimental data, we show that these methods enable robust reconstruction of traction without requiring a difficult selection of regularization parameters specifically for each data set. Thus, Bayesian methods can mitigate the considerable uncertainty inherent in comparing cellular tractions in different conditions.
BibTeX
@inproceedings{huang2019traction,
  author = {Yunfei Huang and Christoph Schell and Tobias B. Huber and Ahmet Nihat Şimşek and Nils Hersch and Rudolf Merkel and Gerhard Gompper and Benedikt Sabass},
  title = {{Traction force microscopy with optimized regularization and automated Bayesian parameter selection for comparing cells}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-018-36896-x},
}

Loading related content...

Figure