Experimental Parameters-Based Monte-Carlo Simulation of Single-Molecule Localization Microscopy of Nuclear Pore Complex to Evaluate Clustering Algorithms

Abstract

Single-molecule localization microscopy (SMLM) enables the detailed visualization of nuclear pore complexes (NPC) in vitro with sub-20 nm resolution. However, it is challenging to translate the localized coordinates in SMLM images to NPC functions because different algorithms to cluster localizations as individual NPCs can be biased without ground truth for validation. We developed a Monte-Carlo simulation to generate synthetic SMLM images of NPC and used the simulated NPC images as the ground truth to evaluate the performance of six clustering algorithms. We identified HDBSCAN as the optimal clustering algorithm for NPC counting and sizing. Furthermore, we compared the clustering results between the experimental and synthetic data for NUP133, a subunit in the NPC, and found them to be in good agreement.