I did that here:

basically, I take an existing dataset and get the mean and std for all probes and look at the percentiles. Obviously, we will have more power if we have a large difference in means with small standard deviations. You can look across a range of assumptions to see how they affect your require sample-size or effect-size.