This one is really interesting and an idea that I like more the more I think about it. The authors capture the idea really well in the first picture.
In general, the idea is this: its hard to find the biomarkers because we see hundreds of thousands or millions of things in a standard peptide ID experiment. So we eliminate a ton of stuff from contention by assessing our entire system variability (sample prep, LC, mass spec, data processing) by comparing two control groups to one another. This gives us a baseline to go on. Then the stuff thats weird in our experimental sample can be considered to have some validity if it exceeds the total variation limit within our experiment.
Now. The big question in my mind is how to I easily do this and automate it. Cause the Qu lab has a pretty great bioinformatician or three...and I don't.... but I'm gonna give this one a whirl later with some commercial tools!