Bimodal protein expression and phosphorylation detected across cancer types associate with known oncogenic processes including EMT.

<p>(<b>A</b>) Two-component Gaussian mixture model fit to E-cadherin protein expression. The lines indicate the probability density contribution from the low (-) and high (+) expression components. The histogram represents the RPPA measurements for the cell lines. (<b>B</b>) By comparing a two- versus one-component fit using the Bayesian Information Criterion (BIC), 260 out of 450 RPPA measurements supported bimodal expression. (<b>C</b>) Heat map of the posterior probabilities of each cell line belonging to the low (-, blue) or high (+, red) mixture component for the top-20 most bimodal proteins. The posterior probabilities can be thought of as soft assignments for the cell lines to low or high expression. Shannon entropy of the tissues assigned to low and high expression quantify the tissue diversity giving rise to the bimodal fits. (<b>D</b>) Overview of classification approach of proteins in terms of bimodality, tissue diversity (Shannon entropy), and frequency of cell lines assigned to the fitted distributions. (<b>E</b>) Significant GO terms for common bimodal proteins that were not found to be significant for non-bimodal proteins (p < 0.05, Benjamini-Hochberg).