The genes are ranked in each sample using the estimated CDFs. This means that for each gene in a sample, the y-axis value for the CDF is used corresponding to the sample x-coordinate.