Fig. 6

Module 3 continued. a Continued overview of this module. After feature extraction (step1, Fig. 5), the dimensionality of the feature maps was reduced, and data points were embedded in a 2D manifold for visualization. Then, cells with similar features were grouped together by clustering. b A 2D embedding by UMAP showing two clusters of data. c ROIs from the smaller cluster were noise and were discarded. d Mean Silhouette Coefficient for all datapoints in the larger cluster, cluster 2, were computed at different cluster numbers. The highest score indicated the optimal cluster number. e Cluster 2 was further partitioned using spectral clustering into four subclusters. For each subcluster, 20 nearest neighbors of the centroid (blue crosses) were visually inspected and their PTEN localizations were manually classified. 10 were plotted, and the predominant PTEN localization for that cluster was indicated