A light-weight, data-driven segmentation method for multi-state Brownian trajectories

Abstract

Single-particle tracking methods have emerged as a crucial tool for the characterization of dynamical and diffusive processes in a range of biological and synthetic systems. Here, we propose a simple and light-weight yet accurate method for the segmentation of multi-state Brownian trajectories based on an optimised Gaussian filtering of the displacement time series combined with an automated fitting to a Gaussian mixture model. We verify our method using synthetic, 2-state Brownian trajectories and show that our method provides high levels of accuracy in terms of segmentation and the estimation of self-diffusion coefficients for reasonably well-separated values of the diffusion coefficients. We furthermore demonstrate the feasibility of our method on experimental systems using single-particle tracking data for diffusing membrane proteins bound to a supported lipid bilayer. Compared to methods based on deep learning or hidden Markov models our method imposes a much lower computational load, making it suitable for fast and accurate online processing of single-particle trajectories from microscopy images.