Fig. 2: Pixie identifies accurate and consistent pixel-level features in lymph node tissue.
From: Robust phenotyping of highly multiplexed tissue imaging data using pixel-level clustering
a Overview of pixel clustering in Pixie. Individual pixels are clustered using a self-organizing map (SOM) based on a set of phenotypic markers. The clusters output by the SOM are metaclustered using consensus hierarchical clustering. If necessary, users can manually adjust the metaclusters, then annotate each metacluster with its phenotype based on its expression profile using our easy-to-use GUI. b Heatmap of mean marker expression of pixel cluster phenotypes for an example dataset of lymph node samples. Expression values were z-scored for each marker. c Multi-channel MIBI-TOF image of a representative field-of-view (FOV) (left), the corresponding pixel phenotype map (middle), and representative insets (right). Colors in the pixel phenotype map correspond to the heatmap in (b). d The FOV in C colored according to the cluster consistency score. e Distribution of cluster consistency score across all pixels in the dataset. f Comparison of cluster consistency score across different pre-processing steps. Boxplots show median as the center and 25th and 75th percentiles as the bounds of the box. n = 12,515,748 pixels from 12 images. **** indicates p value < 2e-16 using a two-sided Wilcoxon test.
