Figure 2

Schematics of FINDER algorithm (a) In DBSCAN, a new cluster is initiated when at least one core point (shown in red) is present that has at least minPts other points within distance \(\varepsilon\) from the core point (see circles, left). Inspired by DBSCAN, the clustering algorithm used in FINDER iteratively removes non-core points (shown in black) which results in a more frequent identification of noise localizations (grey points, right). (b) Two clustering assigments are considered similar, if the number of matching localizations is greater than the number of unmatched localizations. Example of two similar cluster assignments (top row) and two non-similar cluster assignments (bottom row). (c) Phase space of possible clustering outcomes. FINDER computes a similarity score among clustering results sharing the same value minPts (i.e., for each line on the plot, like the one highlighted with the dashed line). (1–3) represent three possible clustering outcomes within the parameter space. The parameters used for (1)-(3) correspond to the location of the respective number in the phase diagram, respectively.