Calculate P-value and FDR-adjusted P-value on normalized data

Question

[Yun]

Hi all,

I just have a juvenile question about calculating P-value and FDR-adjusted P-value on normalized data. My lab did some RPPA analysis way back in 2016 and those data was sitting there until my mentor decided to publish a paper using the data. We sent our data to MD Anderson and they returned to us a bunch of Excel files that only contain normalized log2 and median centered values and all values were normalized for protein loading. However, they did not provide us any RAW file or P-value in the data. Therefore, I am just wondering if there is any method that I can use to calculate P-value while controlling for multiple testing?

Here is a screenshot of the result file. Each condition was run in duplicates. So, we have 5 conditions and 10 samples ran.

Thank you all so much for your help!

Answer 1

[dariober]

However, they did not provide us any RAW file or P-value in the data. Therefore, I am just wondering if there is any method that I can use to calculate P-value while controlling for multiple testing?

I'm not familiar with RPPA but from the quote above I surmise some confusion about p-values.

You first formulate a null hypothesis and then the p-value tells you how plausible is the observed data under the null hypothesis assumptions (not exactly this but I think that's an ok description). So if you don't tell what null hypothesis you want to test you cannot get any p-value. Perhaps you want to test whether there is reason to think that group A is different from group B but if you don't ask for it then your data analyst (MD Anderson) cannot give you any p-value.

Also, typically p-values are not really the final output you want - they are more like an indication of where you should look further. I mean, the purpose of an analysis is not to find small p-values since a small p-value may still be a false positive or an uninteresting result, for example when the null hypothesis or its assumptions are highly unlikely to hold.

Regarding FDR or other multiple testing adjustments, that is usually a simple step once you get your list of p-values for a meaningful test.