Dear all, Apologies if this is off-topic but I have been pondering about multiple correction procedures when multiple groups are involved. In a microarray context, say we have 20000 probes/genes and the responses are either Grade 1, 2, 3 (assume all arrays correspond to some tumor). The null hypothesis is that the gene expression means does not vary between the three levels of the grade. Suppose, I perform a pairwise t-test for these 3 grade on a gene-by- gene basis resulting in 3 p-values for each gene. Do I adjust the p-value within each gene using some multiple correction technique followed adjustment for multiple hypothesis for 20000 probes ? Or just perform the adjustment for multiple probes only ? The other solution I have is to perform F-test, adjust the F-test p-values and select top $n$ genes. Then perform pairwise t-test on the $n$ genes with adjustment to determine which group and how many group means differ. I think this is the more sensible method. My second question is which of these methods better ? Many thanks in advance. Regards, Adaikalavan Ramasamy.

Dear Adaikalavan, Based on your null hypothesis, the F-test seems to be the right thing do. The paired t-tests are more suitable for later investigations once the F-test is rejected. It helps to consider gene and grade as factors. So the only difference between t-tests and F-tests is that when the grade have more than 2 levels, we get an F-test for each gene, instead of a t-test. Yuan Ji, Ph.D. ############################################ Assistant Professor Department of Biostatistics The University of Texas M.D. Anderson Cancer Center 1515 Holcombe Blvd. - Unit 447 Houston, TX 77030-4009 (713)794-4153 ############################################ |---------+------------------------------------------> | | | | | | | | "Adaikalavan RAMASAMY"| | | <ramasamya@gis.a-star.edu.sg>| | | Sent by:| | | bioconductor-bounces@stat.math| | | .ethz.ch| | | | | | | | | 12/15/03 10:10 AM| | | | |---------+------------------------------------------> >------------------------------------------------------------------- ----------------------------------| | | | | | | |To: | | <bioconductor@stat.math.ethz.ch> | |cc: | | | |Subject: | | [BioC] Multiple hypothesis correction and pairwise tests | | | >------------------------------------------------------------------- ----------------------------------| Dear all, Apologies if this is off-topic but I have been pondering about multiple correction procedures when multiple groups are involved. In a microarray context, say we have 20000 probes/genes and the responses are either Grade 1, 2, 3 (assume all arrays correspond to some tumor). The null hypothesis is that the gene expression means does not vary between the three levels of the grade. Suppose, I perform a pairwise t-test for these 3 grade on a gene-by- gene basis resulting in 3 p-values for each gene. Do I adjust the p-value within each gene using some multiple correction technique followed adjustment for multiple hypothesis for 20000 probes ? Or just perform the adjustment for multiple probes only ? The other solution I have is to perform F-test, adjust the F-test p-values and select top $n$ genes. Then perform pairwise t-test on the $n$ genes with adjustment to determine which group and how many group means differ. I think this is the more sensible method. My second question is which of these methods better ? Many thanks in advance. Regards, Adaikalavan Ramasamy. _______________________________________________ Bioconductor mailing list Bioconductor@stat.math.ethz.ch https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor