Fig. 2: Comparison of cluster analysis methods.

a Cluster-rich areas from larger simulated datasets (Supplementary Fig. 1). A fixed number of localizations were split between the designated cluster areas (marked, covering 4% of the total area in the full-size datasets) and non-clustered areas with incrementally increasing ratios of the localization density between the clustered (marked circles) and non-clustered areas, the total number of localizations being constant. Note the ratio of 1 means that the image does not contain areas of deliberate localisation enrichment. b Nearest neighbour analysis for the clustered molecule to itself and to a randomly distributed molecule. ClusA is a protein partitioning to the cluster areas at different ratios. Rand is a random distribution in a flat and smooth membrane. The scale on the x-axis represents the distance in pixels. c Pair correlation analysis of the clustered molecule to itself and a randomly distributed molecule. The scale on the x-axis represents the distance in pixels. d, e A cluster-rich portion from the whole 2048 × 2048 pixel dataset is displayed after Gaussian filtering (sigma 20 pixels). Subtraction of the Gaussian filtered images was performed either with (d) a second Gaussian filtered image of a random distribution (ClusA-Rand) or (e) another image of clustered molecules with the same partition ratios but different absolute distributions both inside and outside the clusters (ClusA-ClusB). ClusB is a membrane marker partitioning at the same ratios to the cluster areas as the protein marker (ClusA). For each ratio (d) and (e) use the same intensity range with zero at the centre and the maximum taken from the peak intensity in the full size image (Supplementary Fig. 1). f Intensity histograms from d (red) and e (black). g The skewness of the intensity distributions of difference image histograms, n = 8 with the line showing the mean. All analyses used the large datasets (Supplementary Fig. 1).