I am doing a GO analysis for my gene sets and plan to implement the Benjamini-Hochberg method to adjust the resulted pValues for multiple testing correction. Since the BH method depends on the total number of testing or pValues calculated, I wonder if it is ok or not to remove all GO terms with only 1 gene hits (or those with 1 or 2 gene hits) before calculating the pValues? In that way, the total number of pValues will be reduces, which may produce more significant adjust pValues. The logic is that the GO terms with just 1 or 2 genes hits are more likely not to be significant.
So my plan is like this:
Is this procedure ok or not? Are there any published papers with similar procedures? Any comments or references will be appreciated. Thank you!
Hi,
It is not ideal to remove those while performing enrichment analysis. But later when you are filtering GO terms, you may consider FDR parameter along with number of genes/hits to be in the term as filter criteria. But when you are calculating FDR, it must contain all the hits and their pvalues, otherwise you create a bias in your analysis. You can have a look at these articles where I considered P-value cutoff along with minimum number of genes in each term as cutoff to filter the terms.
Articles:
https://www.nature.com/articles/s41467-018-03265-1#Sec15
https://academic.oup.com/nar/article/46/18/9384/5053167#122402618
https://clinicalepigeneticsjournal.biomedcentral.com/articles/10.1186/s13148-016-0274-6#Sec2
It looks like you used a threshold of at least 5 genes of 5% of a pathway. How did you decide on those thresholds? Do you have a reference for that minimum size?
Thanks!
Log in to answer this question.
