Combined online Bayesian and windowed estimation of background and signal localization facilitates active-feedback particle tracking in complex environments

Abstract

Despite successes in tracking single molecules in vitro, the extension of active-feedback single-particle methods to tracking rapidly diffusing and unconfined proteins in live cells has not been realized. Since the existing active-feedback localization methods localize particles in real time assuming zero background, they are ill-suited to track in the inhomogeneous background environment of a live cell. Here, we develop a windowed estimation of signal and background levels using recent data to estimate the current particle brightness and background intensity. These estimates facilitate recursive Bayesian position estimation, improving upon current Kalman-based localization methods. Combined, online Bayesian and windowed estimation of background and signal (COBWEBS) surpasses existing 2D localization methods. Simulations demonstrate improved localization accuracy and responsivity in a homogeneous background for selected particle and background intensity combinations. Improved or similar performance of COBWEBS tracking extends to the majority of signal and background combinations explored. Furthermore, improved tracking durations are demonstrated in the presence of heterogeneous backgrounds for multiple particle intensities, diffusive speeds, and background patterns. COBWEBS can accurately track particles in the presence of high and nonuniform backgrounds, including intensity changes of up to three times the particle’s intensity, making it a prime candidate for advancing active-feedback single fluorophore tracking to the cellular interior.