Hello, I am using limma to select differentially expressed genes. I have 24 arrays and 40k genes. According to the limma users' guide, "If none of the raw p-value are less than 1/G, where G is the number of genes, then all of the adjusted p-values will be equal to 1". I get raw p-values which are less than 1/G after applying eBayes; however, the lowest adjusted p-value I get using "fdr" is 0.66. Does that mean that I cannot adjust for multiple testing in experiments involving many genes? Should I then use an arbitrary cut-off on the raw p-values? or what are the alternatives? Thanks! Lourdes

Lourdes Pe?a Castillo wrote: > Hello, > > I am using limma to select differentially expressed genes. I have 24 > arrays and 40k genes. According to the limma users' guide, "If none of > the raw p-value are less than 1/G, where G is the number of genes, > then all of the adjusted p-values will be > equal to 1". > > I get raw p-values which are less than 1/G after applying eBayes; > however, the lowest adjusted p-value I get using "fdr" is 0.66. Does > that mean that I cannot adjust for multiple testing in experiments > involving many genes? Should I then use an arbitrary cut-off on the > raw p-values? or what are the alternatives? The problem here is that you don't have any evidence for differential expression (which is *not* the same as saying there are no differences). With 24 arrays this is sort of hard to believe, unless you have say, 12 samples and only duplicates for each. Regardless, you do have options. First, you can filter the genes prior to doing the statistics to reduce the number of comparisons you are doing. See the genefilter package for various methods of doing this. Second, if you have an a priori idea of the 'type' of genes that you expect to be differentially expressed, you can do the statistical analysis using only those genes. This sort of analysis is a bit more difficult to do than the agnostic filtering done by genefilter - you may need to extract those genes that have a certain gene ontology term associated with them. This may take some work, but is quite doable. Third, even if you don't do one of the above, limma does return genes ranked in likely order of importance. You could simply take the top n genes (where n is dictated by the time and resources you are willing to expend) and try to validate them using qPCR or Northerns, etc. Note that you should use new samples if you want to generalize the results to a population other than the existing samples. Best, Jim > > Thanks! > > Lourdes > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor -- James W. MacDonald Affymetrix and cDNA Microarray Core University of Michigan Cancer Center 1500 E. Medical Center Drive 7410 CCGC Ann Arbor MI 48109 734-647-5623