When using GEO2R to analyze a particular GSExxx (8 samples, 4 ctrl and 4 test, on U133_plus 2 array), the output shows that even for the probe with the lowest p (0.00000265) and the highest logFC (3.9724151), the probe has an FDR or the adj.P.val at 0.114.
Isn't FDR conventionally defined to be significant ONLY WHEN the value is smaller than 0.05 at the minimum? If so, does this imply for this particular GSExxx, ALL probes show no significant differential expression? Or do I misunderstand the meaning/interpretation of FDR or that of adj.p.value under GEO2R analysis?
I would consider a fold-change greater than 1.5 and FDR < 0.05 to be standard criteria for differential expression.
That said, there are a few caveats:
1) You can sometimes see FDR cutoffs up to 0.25
2) Depending upon the journal, you can sometimes get away without the FDR correction. More specifically, a genomics journal will frown on such an action but an article that only uses microarray or high-throughput sequencing data for ~10% of the article will probably not care as much (because that is only one piece of evidence supported by other pieces of evidence).
When I provide gene expression analysis, I always point out the standard criteria. If genes cannot be defined using those criteria, I loosen the criteria but explain that there false positive rate will be higher (and you are guaranteed to see some genes with p < 0.05 by chance).
Validation is important - this is true even for those genes with FDR < 0.05. This is also a potential disadvantage in this specific case: you can't provide any further validation for these specific samples. If you have your own dataset with a more detailed characterization and you are testing the robustness of your results, this may be OK. In that case, the FDR correction may also not be necessary because you only really care about the genes that changed in your other dataset.
