Hi, Here is my question and I hope it fits into the BioC support group (!): I have one cell line A with a deficient gene and another one B with the corrected gene (same cell line indeed with gene transfered). I have some Xpts conducted under normal cell culture conditions and some others in which I have added a stressing agent - some at 10 nM and some others at 100 nM ie 2 different concentrations -. Let say the question is what are the genes that are differentially expressed between A and B ? I was planning to do a first comparison A vs B under normal cell culture conditions, then a second for those ran at 10 nM of the stressing agent, then a third one at the 100 nM dose... But I was wondering if another one, pooling all the A chips versus all the B chips -whatever the stressing agent is added or not and using the average signal- should give me more power to detect some kind of differences because of the increase number of chips in this case ? Said differently I guess it might possible there to pick genes that have not been identified in any previous comparisons just because of the lack of power / not enough chips ? Then in this comparison could I say that I was looking for genes that are differentially expressed between A and B whatever the conditions were, ie with or without drug, means genes that are or seem to be invariantly dysregulated in cells A ? The other approach I see is to do all the first comparisons for each subgroup and to do Venn Diagram Union: under normal cell culture conditions U drug 10 nM U drug 100 nM but then I am not taking the maximum power of the system since I reduce the number of chips per comparison ? It seems that another approach is proposed in T Speed book ie factorial design Xpts but I m waiting for the book ?! thanks for any help / advise Philippe [[alternate HTML version deleted]]

Hi, Sounds to me like a straight-forward balanced 2-way ANOVA unless I'm misinterpreting something. Cell line (A vs. B) is one main effect, stressing agent (0, 10, and 100 nM) is the second, and there is an equal representation of each combination in biological replicates. Significance and multiple testing correction should be assessed with the overall ANOVA p-value (not the main effects and interaction p values), and the main effect and interaction p-values can be considered as post hoc tests (no multiple testing correction necessary once it is controlled for at the overall level). Remember that a significant interaction term trumps main effect terms, basically proving that there is some sort of violation of the underlying assumptions regarding the testing of the main effects (and so genes showing a significant interaction term must be considered separately from those that do not). This seems to be the right approach for this data, but I warn you that the resulting lists can get fairly complicated when you consider all of the combinations of significant main effects, interactions, and pairwise comparisons that can be used to define different sets of genes (e.g., the list of genes that decreased from cell line A to B, and were significantly affected by 100, but not 10, nM stressing agent). HTH, -E >Date: Sun, 20 Apr 2003 10:59:24 EDT >From: Phguardiol@aol.com >Subject: [BioC] methodological question >To: bioconductor@stat.math.ethz.ch >Message-ID: <1e0.72ce9ee.2bd40fcc@aol.com> >Content-Type: text/plain > >Hi, > >Here is my question and I hope it fits into the BioC support group (!): >I have one cell line A with a deficient gene and another one B with the >corrected gene (same cell line indeed with gene transfered). >I have some Xpts conducted under normal cell culture conditions and some >others in which I have added a stressing agent - some at 10 nM and some >others at 100 nM ie 2 different concentrations -. >Let say the question is what are the genes that are differentially expressed >between A and B ? >I was planning to do a first comparison A vs B under normal cell culture >conditions, then a second for those ran at 10 nM of the stressing agent, then >a third one at the 100 nM dose... >But I was wondering if another one, pooling all the A chips versus all the B >chips -whatever the stressing agent is added or not and using the average >signal- should give me more power to detect some kind of differences because >of the increase number of chips in this case ? Said differently I guess it >might possible there to pick genes that have not been identified in any >previous comparisons just because of the lack of power / not enough chips ? >Then in this comparison could I say that I was looking for genes that are >differentially expressed between A and B whatever the conditions were, ie >with or without drug, means genes that are or seem to be invariantly >dysregulated in cells A ? >The other approach I see is to do all the first comparisons for each subgroup >and to do Venn Diagram Union: under normal cell culture conditions U drug 10 >nM U drug 100 nM but then I am not taking the maximum power of the system >since I reduce the number of chips per comparison ? >It seems that another approach is proposed in T Speed book ie factorial >design Xpts but I m waiting for the book ?! >thanks for any help / advise >Philippe