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

Density-based clustering is used in many contexts including in single molecule localization microscopy (SMLM), where it is commonly used to elucidate the nanoscale organization of molecules. However, little guidance is available for evaluating clustering performance, which is often dependent on user-input parameters. Here, we develop an efficient implementation of density-based cluster validation (DBCV) that can quantitatively evaluate clustering in SMLM-sized datasets. We demonstrate that maximizing DBCV scores accurately identifies clusters in noisy, simulated data. Coupling DBCV with Bayesian optimization, we outline a method, DBOpt, that selects input parameters in an unbiased manner for density-based clustering algorithms. We demonstrate that optimal parameters can be selected for popular algorithms (DBSCAN, HDBSCAN, OPTICS) with minimal user input. Finally, we show that DBOpt reports accurate feature sizes in 2D and 3D experimental data. In sum, DBOpt will improve the integrity, reproducibility, and quality of cluster analyses for SMLM and beyond.