Automated Single Molecule Clustering of Superresolution Data as an N-Body Problem

Abstract

Proteins form molecular complexes in cells that are critical for cellular processes such as signal transduction, cellular metabolism, and muscle contraction. Single molecule based super-resolution microscopy methods such as STORM or PALM can resolve the clustered state of these proteins with high spatial resolution. These methods use photoswitching or on/off binding of fluorophores to achieve a sparse image of single molecules in a densely labeled sample. The molecules are localized iteratively over time to build a high-resolution image, which is pointllistic in nature. An automated and robust single molecule clustering algorithm is required to rigorously analyze these single molecule data in a quantitative manner. We propose a clustering algorithm inspired by Symplectic N-Body integrators implemented in modern celestial mechanics simulations. This approach is scalable for big data applications and provides a physical definition for segmented clusters. The performance of this method is compared with a Voronoi based clustering algorithm on simulated and biological data using external and internal validation metrics.