u-track goes 3D!
Particle tracking is ubiquitous in live cell imaging studies. Our lab engaged in the development of particle tracking software first in the context of measuring cytoskeleton polymer dynamics by Fluorescent Speckle Microscopy (Ponti et al. Science, 2004; Yang et al. Nature Cell Biology, 2007; and many more). In 2008, Khuloud Jaqaman extended our tracking capabilities with the u-track algorithm, eliminating the restrictions of the speckle tracking software to smooth and spatially coherent particle trajectories (Jaqaman et al. Nature Methods, 2008). U-track enables the tracing of incoherently moving particles that merge and split and also can disappear for several time points. Over the years other lab members expanded the capacity of the u-track framework. Among the best known expansions is Kathryn Applegate’s development of a motion model for the tracking of growing microtubule plus ends (Applegate et al. J. Struct Biol. 2011). A lesser-known model also implements a mechanism for backwards interpolation of microtubule shrinkages, carefully validated relative to manual ground truth data both in Applegate et al. and in a study with Torsten Wittmann (Matov et al. Nature Methods, 2010), the interpolation approach has been met with skepticism and never got much traction in the community. Another important supplement to the original u-track package was an algorithm invented by Philippe Roudot that enables the tracking of particles with erratic changes in motion (Roudot et al. IEEE Trans Image Processing, 2017).
The computational backbone and the software architecture of u-track was designed for easy generalization of the tracking to 3D. Our own lab made use of this feature for the first time in a study by the Woods Hole Mitosis Consortium (Jaqaman et al. J. Cell Biol. 2010). While several labs followed the example, adoption to 3D has not been widespread as it required some coding expertise. Moreover, many of the more advanced features like the Applegate and Roudot algorithms were not accessible with an easy tweak. With the increasing popularity of light sheet live cell imaging, also in our labs, Philippe Roudot decided to integrally port u-track’s features to a new standalone package called u-track3D (Roudot et al. Cell Report Methods, 2023 ). The paper shows several tracking applications and consensus validations, including once again the backward interpolation of microtubule plus tip trajectories, now in 3D, to infer microtubule disassembly dynamics. Philippe also used the ground truth data from the particle tracking challenge (Chenouard et al. Nature Methods, 2014) to demonstrate that even after 15 years u-track’s trajectory construction remains among the best-performing algorithms, including recent methods relying on Deep Learning. We interpret this outcome as demonstration of the near-optimality of u-track’s two-step solution of the spatial and temporal association problem benefiting from the strong priors built into the underlying bipartite graph matching. It takes a lot for Deep Learning to converge to this level of accuracy without priors. The most innovative parts of u-track3D are, however, not the tracking. Two major challenges with 3D image analysis are visualization and validation of the results. The complexity of following trajectories of individual particles in dense 3D particle clouds by visual inspection is overwhelming in most cases. U-track3D offers tools to address this. Reviewed in a News and Views article by Lance Xu and Steve Presse, u-track3D provides dynamic ROIs (regions of interest) that allow the visualization of a subset of particles in a frame of reference that follows the movements under a certain perspective. Second, the package includes a module that employs stochastic resampling to test the stability of the bipartite graph association (figure below). This stability is determined by the local density of particles and their motion relative to one another. The test is translated into a trackability score that determines at every time point for every trajectory how trustworthy the extracted particle path is. The score can be used in postprocessing steps to filter trajectories based on quality.
The publication and release of u-track3D closes a long-open chapter in our research program. We will no longer work on particle tracking but become users of our own software and software written by others. We will maintain and bug-fix the u-track3D release on github under the auspices of the UTSW-UNC Center for Cell Signaling Analysis.
Philippe Roudot may continue to innovate tracking in his own lab in Marseilles. We recommend tagging his website for updates on tracking technology.
Figure: u-track 3D employs stochastic resampling to assess the particle-trajectory assignment stability and to derive a trackability score.