Anisotropic DBSCAN\nfor 3D SMLM Data Clustering

Single-molecule localization\nmicroscopy (SMLM) advanced\nbiological\ndiscoveries beyond the diffraction limit. Various implementations\nenable 3D SMLM to reconstruct volumetric cell images. Yet, the inherent\nanisotropic point spread function of optical microscopes often limits\nthe localization precision in the axial direction compared to the\nlateral precision. Such localization anisotropy could also expand\nspherical cellular structures to ellipsoidal cellular structures.\nStructure identification, however, is often performed using DBSCAN\ncluster algorithms, considering an isotropic search volume. Here,\nwe show that an anisotropic DBSCAN search volume identifies anisotropic\nclusters more reliably using simulated ground truth data sets. Given\nexperimental localization precisions, we suggest optimized search\nparameters based on an expanded computational grid search and show\nan enhanced performance of anisotropic DBSCAN amidst variations in\nlocalization precision. We demonstrate the capability of anisotropic\nDBSCAN on experimental data and anticipate that the algorithm allows\nfor a more rigorous identification of clusters in cells, considering\nthe anisotropic localization precisions of astigmatism-based 3D SMLM.