Comparison of randomly distributed (simulated) localisation events, platelets of different morphologies and the CD41 protein distribution in a reconstructed dSTORM image.

(a) shows the distribution of CD62p labeled with Alexa647-antibody. (b) represents a randomly distributed point cloud (approx. 85 000 data points/image). (c) shows the comparison of the cluster densities distribution to Ripley’s K-function distributions for all cluster dimensions between the simulated and CD62p datasets (black/blue line for cluster density/K-function, respectively). The lines depict the aggregated p-value of KS- and WX-test results. The test based on the Ripley’s K-function classifies the samples as similar (similar for cluster sizes ranging from 5 nm to about 400 nm). In contrary, our classification estimates the samples as dissimilar, with the test based on the density distribution of clusters (similarity measure = 0). The MLP neural network classifies samples as dissimilar with a posteriori probability of 88%. (d) and (e) show a simulated polarized (randomly distributed) and unpolarised image respectively. (f) shows the comparison of clusters density/curvature distributions and Ripley’s K-functions for all cluster dimensions between the simulated polarized and simulated unpolarized datasets (black/blue line for cluster density/K-function, respectively). The lines depict the aggregated p-value of KS- and WX-test results. The test based on the Ripley’s K-function classifies the samples as dissimilar (similar for cluster sizes ranging from approx. 150 nm– 180 nm). Our classification estimates the samples as dissimilar, with the test based on the density distribution of clusters (similarity measure = 0). Three additional pairs of polarized and unpolarised samples have been compared, rendering a dissimilarity of 78.5%, 88.5% and 95.1%. (g) and (h) show reconstructed dSTORM images of Alexa647-phalloidin labelled actin skeleton of two individual platelets activated on a glass surface. The two platelets are comparable, in an early activation state. (i) depicts the comparison of the cluster densities and the Ripley’s K-function distributions for all cluster dimensions between the datasets representing platelet cytoskeleton at a similar activation stage (black/blue line for cluster density/K-function respectively). The lines depict the aggregated p-value of KS- and WX-test results. The test based on Ripley’s K-function classifies samples as strong dissimilar. Our analysis provides a correct classification of samples as similar, based on the density distribution of clusters (similarity measure = 0.86). The MLP neural network classifies samples as similar with a posteriori probability of 87%. (j) and (k) show reconstructed 3D dSTORM images of Alexa647-phalloidin labelled actin skeleton of two individual platelets activated on a glass surface. The two platelets are in two different activation states, (j) a late activation state and (k) an early activation stat. (l) shows the comparison of the cluster densities and of Ripley’s K-function distributions for all cluster dimensions between the datasets representing localisation microscopy images of platelets in early and late activation state (black/blue line for cluster density/K-function respectively). In this case both tests based on the Ripley’s K-function as well as on the cluster density distribution correctly classify the samples as dissimilar (similarity measure = 0.1). The MLP neural network classifies rather samples as dissimilar with a posteriori probability of 64%. The results clearly prove that our method is also capable to compare protein distributions that change over time induced by external factors at a single cell level. The axial range is represented by two colors: blue is below the focal plane and yellow above. The axial range is ± 500 nm.

(TIF)