Abstract
Multi-wavelength single-molecule fluorescence colocalization methods allow elucidation of complex biochemical reaction mechanisms. However, analysis of colocalization data is an intrinsically challenging and time-consuming problem for multiple reasons. First, to minimize dye photobleaching, images frequently are collected at low signal-to-noise ratios, making it challenging to discriminate real fluorescent spots from noise. Second, transient non-specific interactions of the binder molecule with the surface of the microscope slide can give rise to both false-positive and false-negative detection. Third, current analysis methods require subjective choice of user-set thresholds for such spot parameters as amplitude, diameter and proximity. To overcome these difficulties, we developed a new analysis method based on statistical modeling of the image data. This method: 1) maximizes extraction of useful information from data by analyzing 2-D images, not integrated intensity; 2) discriminates authentic fluorescence spots from fluctuations in background fluorescence in a probabilistic manner and assigns spot probabilities (not merely a binary "spot/no spot" classification); 3) uses a realistic pixel intensity model that accounts for background, spot, and camera noise sources; 4) explicitly models non-specific interactions of binder molecules with the slide surface; and 5) can globally fit data from experimental samples together with negative controls that lack target molecules. The software is implemented in the Python-based Pyro probabilistic programming language which allows easy substitution of different statistical models, efficiently uses parallelized processing (GPU) hardware, and is scalable to large data sets. The algorithm is effective without manual parameter tweaking on both simulated and experiment-derived data sets with a wide range of signal, noise, and non-specific binding characteristics. We anticipate that this work will increase the accessibility and utility of single-molecule fluorescence colocalization methods.