Bayesian optimized parameter selection for density-based clustering applied to single molecule localization...

Area of Science:

• Biophysics
• Computational Biology
• Data Science

Background:

• Density-based clustering is vital for analyzing nanoscale organization in single molecule localization microscopy (SMLM).
• Current methods lack robust guidance for evaluating clustering performance and selecting optimal user-defined parameters.
• This limitation hinders the reproducibility and accuracy of SMLM data analysis.

Purpose of the Study:

• To develop an efficient and quantitative method for validating density-based clustering in SMLM datasets.
• To introduce an automated approach for unbiased selection of optimal parameters for density-based clustering algorithms.
• To enhance the integrity, reproducibility, and quality of cluster analyses in SMLM and other fields.

Main Methods:

• Implementation of an efficient density-based cluster validation (DBCV) algorithm for SMLM-sized datasets.
• Coupling DBCV with Bayesian optimization to create the DBOpt method for automated parameter selection.
• Testing DBOpt with popular algorithms like DBSCAN, HDBSCAN, and OPTICS on simulated and experimental 2D/3D data.

Main Results:

• Maximizing DBCV scores effectively identified clusters in noisy, simulated SMLM data.
• DBOpt successfully selected optimal parameters for DBSCAN, HDBSCAN, and OPTICS with minimal user input.
• DBOpt accurately determined feature sizes in both simulated and experimental 2D and 3D SMLM data.

Conclusions:

• DBOpt provides an efficient and unbiased approach for parameter selection in density-based clustering.
• This method significantly improves the reliability and reproducibility of cluster analysis for SMLM data.
• DBOpt is a valuable tool for enhancing quantitative analysis in SMLM and potentially other scientific domains.