HDCluster: High-Degree Graph Clustering for Robust Analysis of Single Molecule Localization Microscopy.

Area of Science:

• Data Science
• Computational Biology
• Biophysics

Background:

• Clustering is essential for data analysis, grouping similar objects.
• Super-resolution single molecule localization microscopy (SMLM) generates large datasets requiring efficient clustering.

Purpose of the Study:

• Introduce HDCluster, a novel high-degree graph-based clustering algorithm.
• Address the need for rapid and effective clustering in big-data applications, especially SMLM.

Main Methods:

• Developed HDCluster, a graph-based algorithm utilizing high-degree connections.
• Employed a single parameter for flexibility, enabling automatic cluster number determination.
• Validated against state-of-the-art methods on simulated, public, and real-world SMLM datasets.

Main Results:

• HDCluster demonstrated superior time efficiency and clustering performance (ARI, AMI) compared to existing algorithms.
• The algorithm effectively handles large SMLM datasets with millions of localizations.
• HDCluster exhibits robustness to noise, a critical factor in SMLM data.

Conclusions:

• HDCluster is a fast, accurate, and robust clustering tool for big-data challenges.
• It offers a significant advancement for SMLM data analysis and other large-scale clustering tasks.