Dear Loren, > Date: Wed, 05 May 2010 09:28:45 -0700 > From: Loren Engrav <engrav at="" u.washington.edu=""> > To: rbioc <bioconductor at="" stat.math.ethz.ch=""> > Subject: [BioC] Romer warning serious? and nrot=9999? > > So ran romer with > romer(iset, y, design, contrast, array.weights=NULL, block=NULL, > correlation, floor=FALSE, nrot=9999) > > Finished with no errors but I see > > Warning > In romer (iset, y, design, : > Estimation of var.prior failed ? set to default value > > romer2 finished with same message Unlikely to be a problem. It does suggest though that your dataset is relatively small or a bit unusual. > Am I in trouble? > Or can I ignore this? > > Also > Did nrot=9999, is 9999 really necessary? Or is 1000 good enough? The smallest possible p-value is 1/(nrot+1). So nrot=999 (or 1000) means the smallest possible raw p-value is 0.001. This may not be very convincing if you have tens of thousands of sets. Personally, I use nrot=9999 or greater. Best wishes Gordon

Thank you Small data set? Yes, my data set is small, 1842 genes of interest at week1 2 3 12 or 20 between Duroc and Yorkshire breeds after filters And the warning occurs only for week1 where the differential is the smallest of all 5 time points with only a handful of differentially expressed genes 9999? I can run 9999 but the variance in this complex biological system is large so I probably cannot claim .0001 precision, and we are "screening" at this point but if something "interesting" appears I will check that out with 9999 Thank you > From: Gordon K Smyth <smyth at="" wehi.edu.au=""> > Date: Fri, 7 May 2010 10:27:36 +1000 (AUS Eastern Standard Time) > To: Loren Engrav <engrav at="" u.washington.edu=""> > Cc: rbioc <bioconductor at="" stat.math.ethz.ch=""> > Subject: [BioC] Romer warning serious? and nrot=9999? > > Dear Loren, > >> Date: Wed, 05 May 2010 09:28:45 -0700 >> From: Loren Engrav <engrav at="" u.washington.edu=""> >> To: rbioc <bioconductor at="" stat.math.ethz.ch=""> >> Subject: [BioC] Romer warning serious? and nrot=9999? >> >> So ran romer with >> romer(iset, y, design, contrast, array.weights=NULL, block=NULL, >> correlation, floor=FALSE, nrot=9999) >> >> Finished with no errors but I see >> >> Warning >> In romer (iset, y, design, : >> Estimation of var.prior failed ? set to default value >> >> romer2 finished with same message > > Unlikely to be a problem. It does suggest though that your dataset is > relatively small or a bit unusual. > >> Am I in trouble? >> Or can I ignore this? >> >> Also >> Did nrot=9999, is 9999 really necessary? Or is 1000 good enough? > > The smallest possible p-value is 1/(nrot+1). So nrot=999 (or 1000) means > the smallest possible raw p-value is 0.001. This may not be very > convincing if you have tens of thousands of sets. Personally, I use > nrot=9999 or greater. > > Best wishes > Gordon