Tweet
Hi everybody!
I have some questions about the FDR results in baySeq. I read the “R script” for baySeq documentation (pdf file by bioconductor) and loaded the corresponding simulated data (count data of one thousand counts with the first hundred counts having differential expression between the first five libraries and the second five libraries.). Then I followed the steps and estimated empirical distribution (getPriors) and posterior likelihoods (getLikelihood), obtaining the same results as in the example. I used topCounts function for see the differential expression, that show the top 10 candidates with un FDR between 0.001 and 0.005. But when I wanted to see the FDR for the first hundred counts (for all counts that should show differential expression) the FDR goes from ~0.001 to ~0.360, which is puzzling me… How one can consider an FDR that high as “acceptable”? I followed the same protocol with my data (a transcriptome of 66000 transcripts); results says that I have 2% (~ 1300 transcripts) of DE transcripts (a number that agrees with the results using cuffdiff), but the corresponding FDR is about 0.5 which seems to me too high…
Anyone else has the same type of results? Which FDR would you recommend as threshold to consider DE transcripts? If any one could share some ideas I would be thankful.
I have some questions about the FDR results in baySeq. I read the “R script” for baySeq documentation (pdf file by bioconductor) and loaded the corresponding simulated data (count data of one thousand counts with the first hundred counts having differential expression between the first five libraries and the second five libraries.). Then I followed the steps and estimated empirical distribution (getPriors) and posterior likelihoods (getLikelihood), obtaining the same results as in the example. I used topCounts function for see the differential expression, that show the top 10 candidates with un FDR between 0.001 and 0.005. But when I wanted to see the FDR for the first hundred counts (for all counts that should show differential expression) the FDR goes from ~0.001 to ~0.360, which is puzzling me… How one can consider an FDR that high as “acceptable”? I followed the same protocol with my data (a transcriptome of 66000 transcripts); results says that I have 2% (~ 1300 transcripts) of DE transcripts (a number that agrees with the results using cuffdiff), but the corresponding FDR is about 0.5 which seems to me too high…
Anyone else has the same type of results? Which FDR would you recommend as threshold to consider DE transcripts? If any one could share some ideas I would be thankful.