Dear all,
We got RPPA data (~300 proteins) from a company and I wanted to extract proteins significantly different from the control. We got data already normalized and just needed to extract "differentially expressed" entities, get p-values, etc. Without having a second thought, I used limma, as I have done a number of times with gene microarrays before - results did not look suspicious.
Since then I have been getting questions about the validity of using limma with RPPA, with people pointing out at virtual absence of published RPPA analysis using limma. I could not really find any published analysis of this kind myself. I do not see where limma could fail. As far as I understood it does not even have any assumptions on the underlying signal distribution. Actually, I got some doubts about the latter after seeing G.Smyth's post somewhere about being cautious when using limma with protein counts.
Anyway, long story short, could anybody, please, comment on using limma for extracting "DE'd" species in RPPA (I also have a smaller RTK to analysis of which I wanted to apply it). Is there a problem with too few species in case of ~300 or ~50 species? Thank you
limmadoes assume normally distributed data. I have no idea if RPPA data fit the assumption.However, I found two BioConductor packages for analyzing protein arrays: RPPanalyzer and PPA. You could either drop limma and use one of these packages in its place, or use them to validate your limma analysis - bioinformatics methods very often give different "good" results, though.