"Believe it or not: how much can we rely on published data on potential drug targets?" [Link]
which investigates the reproducibility of some drug models and surveys. And for example, 43 out of 66 data sets were inconsistent!
Another survey shows, that three different, trained individuals assign different states in the cell cycle to a picture strip of a dividing cell. Okay, indeed, they're different people, so that's absolutely fine, but even one person diverges from its own former choice by about 4% when they decide about the cell cycles again. Here's the link.
What's the point? A solution as suggested in the paper I just outlined is the simulation and "calculation" of the points when the cell cycle begins the next "step". Therefore, an objective criterion is necessary - and that's one of the things we will take care of in "Computational Systems Biology". In general, it's quite useful to have a model of your experiment: You don't have to carry out everything, so you save resources, money, time, ... and you could even verify your theory, model a situations you can't build in the lab etc.
Thus, only trust the survey you falsified yourself!
