Fig. 3: Structure-aware tracking improves tracking error and resolution.

a Principle of structure-aware tracking: fluorescence image of the organelle structure (left), segmented image (middle), and the derived graph representing the local pixel connectivity (right) from which the distance between two points along the structure is approximated as the shortest path between their nodes in the graph. Examples of simulated trajectories where structure-incorrect displacements (with graph but not Euclidean distances greater than the linking distance) are displayed in red for an ER (b, left), mitochondria (b, right) or parallel tubule (c) geometries. Ratios of Fidelity (d) or Ambiguity (e) scores obtained with structure-aware tracking over conventional tracking at their respective optimal linking distances for a wide range of characteristic lengths and spot densities for the parallel-tubule simulations. f Total Internal Reflection Fluorescence (TIRF) image of entangled neurites of neurons derived from human induced pluripotent stem cells expressing a cytosolic HaloTag stained with PA-JF646 dye. The image was acquired post-SPT recording with high 405 nm laser power to visualise the marker in bulk. The overlay shows the contour of the identified neurites. g Single-particle trajectories data obtained with conventional tracking of the cytosolic probe for the field of view presented in (f) and overlaid with the contour of individual neurites. The insets show two different incorrect trajectories spanning multiple neurites. h The same single-molecule recording as in panel (g) this time tracked with structure-aware tracking. i Percentage of ambiguous displacements in the conventional and structure-aware tracked datasets. j Same as (d) but for simulations in ER (top) or mitochondria (bottom) geometries. k Same as (e) but for simulations in ER (top) or mitochondria (bottom) geometries. l Ratios of the densities of recovered non-ambiguous displacements as a function of the simulated spot density when computing ambiguities with the graph distance over computing ambiguities with the Euclidean distance, for different characteristic lengths (colours) for simulations in ER (top) or mitochondria (bottom) geometries. The results derived from simulations are averages over 3 and 5 independent repeats for the parallel-tubule and for each of the 6 ER and 5 mitochondria geometries, respectively. Source data are provided as a Source Data file.