All the data we cannot see: maximizing information extraction from fluorescence localization imaging with photobleaching using reversible jump MCMC and hidden Markov modeling

Fluorescence localization imaging with photobleaching (FLImP) is a super-resolution technique developed to automate single-molecule imaging of membrane proteins at sub-10nm resolutions. FLImP combines one-dimensional photobleaching traces with fluorophore point-spread functions to produce molecular separation fingerprints. This enables detailed characterization of the spatial organization of membrane proteins such as the epidermal growth factor receptor (EGFR), where DNA mutations can alter oligomerization and contribute to cancer development. A single FLImP series can generate up to 10,000 integrated intensity traces (tracks), yet lesser than 1% of these typically meet the quality thresholds for downstream analysis, owing to a heuristic “20-questions” track selection process. This inefficiency means FLImP currently requires imaging 1,000s of cells to construct molecular separation fingerprints with <3nm resolution and so limits clinical translation, where sample sizes are limited to fewer than 100 cells. To overcome this problem, we introduce a statistically grounded analysis pipeline designed to maximize information extraction by considering the number and locations of photobleach and “off”-state events as a change point problem in a biologically motivated reversible jump Markov chain Monte Carlo approach, which targets short-lived blinking and dark states. This is followed by post-processing track selection that combines hidden Markov modelling and per-frame filtering to improve the labelling of individual fluorophores and the overall track output quality. Simulated and experimental FLImP data are used to validate the method’s performance, demonstrating improved accuracy and data retention when compared to leading alternatives. This enhanced pipeline directly supports the clinical translation of FLImP by making analysis of small, patient-derived samples, such as circulating tumor cells, feasible, and paves the way for the development of diagnostics and targeted therapies through patient stratification based on protein oligomerization profiles.