Bioconductor Rankprod Strange Results.

Question

[Iddius]

Hi everybody, I'm currently working on a project which involves the measurement of genome-wide gene expression through deep-sequencing. We calculate the gene expression as RPKM as described in literature and then we apply a rank product analysis to get the differentially expressed genes. To apply the rank product statistic we use the Bioconductor RankProd package. Everything is fine except for one dataset (two replicates, for two conditions). When calling the topGene function and imposing a cutoff of 0.05, we have as a result an unfiltered list of genes. In particular in the list we find many genes with a pfp > 0.05. Unfortunately, I can not figure out what is wrong. Any ideas? The dataset can be find here

Thanks for your help.

[Ying W]

Could it be possible that there are more differentially expressed genes between those two conditions? If you are using the same code/workflow and it gives expected results for other datasets then this could be a conclusion. If you are doing something different code-wise in this dataset then maybe it is a coding issue? What do you mean by "unfiltered list of genes"?

Answer 1

[Luca Beltrame]

It looks like a bug some RankProd versions: I experienced it myself several months ago. As a workaround, just do the filtering manually (topGene doesn't do much more than reassembling and filtering the results from the RP test):

rp.tables <- topGene(myRP.result, cutoff=0.05, method="pfp", logged=T, logbase=2)

up.genes <- rp.tables$Table1[rp.tables$Table1$pfp <= 0.05, ]
down.genes <- rp.tables$Table2[rp.tables$Table2$pfp <= 0.05, ]