Fig. 5
Analysis of intra-sample heterogeneity using IF. (a) Multi-scale deconvolution of nuclear pixel intensity variation. Top: pS6 intensity for a tumor section. Bottom: breakdown of the original pattern as a sum of expression at different length scales, from whole section to subcellular. (b) Variation for individual biomarkers (columns) across different spatial scales (rows). Grayscale color: percentage of variance in biomarker pixel intensity captured across length scales (columns). Sample images are shown for high vs low (solid vs. dashed orange line) variation at different length scales. (c) Confidence that sub-samples from a PDX model capture full model heterogeneity depends on the sampling strategy. Panels: results for images from models PDX-L1 and PDX-L2 respectively, stained for IF marker set 1. Results are averaged over 40 PhenoRipper models. Y-axis: confidence that PhenoRipper profiles of a collection of sub-images “matches” the full model profile. X-axis: number of sub-images in collection. Curves: different strategies of selecting sub-images within a model. (d) Box-Whisker plot for the saturation (i.e., highest) confidence level for each sampling strategy across all combinations of models and marker sets.
