Fig. 7

Assessment of protein co-expression in the epithelium and stroma by computational image analysis. We performed a large-scale validation experiment of predicted cancer self-loops by evaluating 1147 images from a common set of 105 proteins in normal breast and breast cancer. We then performed machine learning-based epithelial-stromal segmentation followed by quantitation of protein expression in the epithelium and stroma. Red indicates epithelium and green indicates stroma. Pixels whose class was either unknown or which did not belong to either of the classes are represented in black. After this, pixels containing brown stain in each region were extracted by applying a threshold to the intensity values in the red, green, and blue channels of the image. Brown pixels belonging to epithelium or stroma were reported and analyzed for validation. The analysis shows significantly increased epithelial-stromal protein co-expression in breast cancer as compared with normal breast, as predicted by our network analysis (p < 2.2e-16). There was a trend for increased epithelial-stromal co-expression for predicted self-loops within the cancer samples, although this did not reach statistical significance (45 % vs 38 %, p = 0.13). non-SL non-self-loop; SL self-loop