well, I was doing a differential protein profiling and looking for proteins which significantly up/down-regulated for my treatment. However, I just don't know how I should define "significant". There are some proteins showed changes (p1.3 or

Hi, Chun, I agree with Temba that the analysis one want to carry out is more often than not dictated by the hypothesis laid out to unravel a scientific query. what is it that you want to find out from the design of the experiment, the data collected and the means(statistical tool) to bridge the knowledge gap is the most critical question to help you make an informed decision.

Regards,

Just some concrete points - as the others mentioned, what you actually do of course depends on the experiment design and aim.

Regarding the averaging, if you are reporting average ratios from replicates you should use the geometric mean for the overall trend. This incidentally corresponds to what the Protein Pilot documentation calls the average in log space; it is explained in some detail for instance in the ProteinPilot Online Help document why that is more appropriate.

Regarding missing values, there is always some lack of overlap between runs; you'll never have exactly the same proteins identified even in repeated runs of the same sample. So you can come up with a heuristic that is relevant to your experiment, such as for example use proteins that appear in 2 out of 3 replicates. There is no set rule, you have to decide what is meaningful in your case.

You must check the variability at the instrumental level, after the variability between the technicians. If they agree at the p=0,05 then you can average then. Also keep in mind that you must have enough degrees of freedom at the technical and instrumental levels to reach a meaninful statistical conclusion