Fig. 1.

The concept of diffusional fingerprinting for analyzing and classifying molecular identity based on SPT data. (A) A typical input consists of SPT data obtained by tracking particles in a recorded movie, visualized here as the horizontal planes in the cartoon. Zoom-in: Three typical trajectories (note the different diffusional behaviors). (B) Each trajectory is analyzed and 17 descriptive features underlying SPT diffusional behavior are extracted. The feature values are shown with a gray color code in the horizontal lines of the matrix; these values contain information on the confinement effects, state-shifting diffusion, anomalous diffusion, and non-Brownian displacements. The procedure is repeated for all particle types and conditions, as shown by the color next to the feature matrix. (C) The diffusional fingerprint is composed of the combined feature distributions for each particle type, here shown as a dimensionality-reduced plot, where the surfaces encapsulate 1$\sigma$ of the data points. The diffusional fingerprint of each variant contains information on all observed trajectories. New, unknown trajectories are classified with high accuracy in terms of the known fingerprints, using a simple logistic regression model. (D) Ranking of features offers deconvolution of the most relevant differences between fingerprints and gives key mechanistic insights into diffusional differences between measured conditions and particles.