Hello all, I have a question while I am using the limma package to identify differentially expressed genes: should I perform gene filtering after normalization to exclude genes that are likely unexpressed in the samples before fitting the linear model. With my limited stats knowledge, I believe the inclusion of 'unexpressed' genes may affect the BH mutliple testing correction by unnecessarily increasing the number of genes being tested. Previously when I performed global test (using the globaltest package) on Affy data, however, I found that the gene filtering step had no noticeable effect on the final P-value and therefore had not been required, so I wonder if limma's capability to detect differentially expressed genes would be affected by whether or not 'unexpressed' genes were filtered out. Thanks very much in advance. Fred Peng

Hi Fred, You are correct that including a bunch of unexpressed genes when adjusting for multiplicity will reduce power. However, with limma you don't want to remove the 'unexpressed' genes too early (this doesn't apply to 'bad' data, where the spots are demonstrably unreliable for some reason or another). You have to remember that the eBayes() step adjusts the denominator of the t-statistic based on a prior variance estimate that is calculated from all the genes under consideration. If you filter out genes prior to this step you can bias this estimate. So the recommended method is to perhaps remove demonstrably bad spots first, do the normalization, model fitting, etc, and then filter out those genes you consider unexpressed before doing the multiplicity adjustment. Best, Jim Peng, Fred wrote: > Hello all, > I have a question while I am using the limma package to identify > differentially expressed genes: should I perform gene filtering after > normalization to exclude genes that are likely unexpressed in the samples > before fitting the linear model. With my limited stats knowledge, I believe > the inclusion of 'unexpressed' genes may affect the BH mutliple testing > correction by unnecessarily increasing the number of genes being tested. > Previously when I performed global test (using the globaltest package) on > Affy data, however, I found that the gene filtering step had no noticeable > effect on the final P-value and therefore had not been required, so I wonder > if limma's capability to detect differentially expressed genes would be > affected by whether or not 'unexpressed' genes were filtered out. > Thanks very much in advance. > Fred Peng > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor -- James W. MacDonald, M.S. Biostatistician Douglas Lab University of Michigan Department of Human Genetics 5912 Buhl 1241 E. Catherine St. Ann Arbor MI 48109-5618 734-615-7826