Characterizing cellular receptor dynamics in live fluorescence imaging using stochastic modeling and machine learning

Cell surface receptor dynamics regulate signaling and endocytosis and can be quantified by single-particle tracking using total internal reflection fluorescence microscopy. We developed an interpretable machine-learning framework that combines geometric features of trajectories with a random forest classifier to distinguish five motion classes: Brownian motion, directed motion, fractional Brownian motion, Ornstein-Uhlenbeck motion, and continuous-time random walk. Applied to trajectories of the HIV co-receptor C-C chemokine receptor type 5 (CCR5), the method shows that stimulation with the agonist PSC-RANTES shifts receptor dynamics from intermittent and correlated diffusion toward confined, attraction-driven motion, consistent with receptor clustering and internalization. By providing accurate classification together with interpretable descriptors of receptor behavior, this framework links stochastic motion models to biologically meaningful cellular states. The approach enables quantitative analysis of complex receptor dynamics and is broadly applicable to live-cell imaging studies of membrane proteins and other single-particle tracking datasets.