Dear list, I'm analysing my full factorial experiment in R / Maanova and am having some trouble using "matest" to evaluate which genes show a significant interaction effect, and in particular how to use the contrast matrix to compare specific treatment groups with one another. I hope this is the right place to ask, and I apologise if this has been asked before. I'm analysing 46 samples hybridized to custom designed one-colour Nimblegen arrays. I have two factors: "temperature" (2 levels; 20 or 25) and "age" (3 levels; 20, 50 or 90), in a full factorial design, with 7 to 8 biological replicates and (at this stage) no random variables. My response variables are 15,830 probe-summarised, quantile normalised expression values (log scale). For details and session info see below the email. I'm interested in which genes show a significant temperature x age interaction, and of those genes, which genes show, for each of the three age levels separately, a significant effect of temperature. Of the genes that do not show a significant interaction, I want to know whether the main effects are significant, i.e. which genes show an overall effect of temperature or age. This is my model: >fit1<-fitmaanova(mydata,formula=~temperature+time.point+temperature:t ime.point) the labels of the treatment groups are as follows: > fit1[10] $`temperature:time.point.level` [1] "20:20" "20:50" "20:90" "25:20" "25:50" "25:90" question 1: Am I coding my variables correct? Or should I make a new variable 'treatment group' that with all 6 combinations of the 2 biological factors temperature and age? (as is done in the R / maanova tutorial) Then I want to test the interaction term: > int<-matest(mydata,fit1,term="temperature:time.point", test.type = "ftest", test.method=c(1,1,1,1), n.perm=2000) I know how to access F and p values in int$F1 and int$Fs, but I'm not sure exactly how to interpret them. The contrast matrix int$Contrast suggests that a series of pairwise comparisons was performed: > int$Contrast [,1] [,2] [,3] [,4] [,5] [,6] [1,] 1 0 -1 -1 0 1 [2,] 0 1 -1 0 -1 1 question 2: Is this matest giving me indeed the F and p values for the interaction? Or for the two contrasts specified in int$Contrast? For univariate analyses, I normally use anova(lm(y~factor1+factor2++factor:factor2)) to obtain the F and p values for main effects and interaction in case of such a two-way model. Ideally, I'd like to use p values for the interaction (with a cut-off of, say, 0.1) to divide the set of the genes in two: genes with and genes without a significant interaction effect. For the first group I'd like to test, for each of the three age levels separately, which genes show a significant effect of temperature. For the second I'd like to look at the two main effects. Question 3: is this a sensible approach? Or should I always include all genes in the analyses, and only afterwards compare the genesets of the different tests? E.g. intersect the set that has a significant temperature x age interaction with a set that shows a difference between the 2 temperatures at the first age class? What is the best way to compare the two temperatures within each of the three age classes? With the full data set, I continued to analyse, for each of the three age levels separately, which genes show a significant effect of temperature. I tried to make a contrast matrix C with one row for each the contrasts I care about: > C [,1] [,2] [,3] [,4] [,5] [,6] [1,] 0 0 1 0 0 -1 [2,] 0 1 0 0 -1 0 [3,] 1 0 0 -1 0 0 However, when I use matest with this contrast matrix > int<-matest(e,fit1,term="temperature:time.point", Contrast = C, + test.type = "ttest", test.method=c(1,1,1,1), n.perm=10) It gives this error: Error: The number 1 test is not estimable > traceback() 3: stop(paste("The number", i, "test is not estimable"), call. = FALSE) 2: checkContrast(model, term, Contrast) 1: matest(e, fit1, term = "temperature:time.point", Contrast = C, test.type = "ttest", test.method = c(1, 1, 1, 1), n.perm = 10) I understand that I'm not testing both temperature and age in each contrast (each row of the matrix), and therefore perhaps should not use "temperature:time.point" as term in matest. But I'm not sure how to compare the two temperatures within each of the three age classes. Thanks a lot in advance for any input. Cheers, Vicencio > sessionInfo() R version 2.11.1 (2010-05-31) i386-pc-mingw32 locale: [1] LC_COLLATE=English_United States.1252 [2] LC_CTYPE=English_United States.1252 [3] LC_MONETARY=English_United States.1252 [4] LC_NUMERIC=C [5] LC_TIME=English_United States.1252 attached base packages: [1] stats graphics grDevices utils datasets methods base other attached packages: [1] lattice_0.18-8 maanova_1.20.0 Biobase_2.8.0 limma_3.4.1 loaded via a namespace (and not attached): [1] grid_2.11.1 tools_2.11.1 > mydata Summary for this experiment Number of dyes: 1 Number of arrays: 46 Number of genes: 15830 Number of replicates: 1 Transformation method: None Replicate collapsed: FALSE > table(mydata$design[,18:19]) time.point temperature 20 50 90 20 8 8 8 25 8 7 7 > fit <-fitmaanova(mydata,formula=~temperature+time.point+temperature: time.point) > int <-matest(mydata,fit1,term="temperature:time.point", test.type = "ftest", test.method=c(1,1,1,1), n.perm=2000) > names(fit1) [1] "probeid" "yhat" [3] "S2" "G" [5] "temperature" "temperature.level" [7] "time.point" "time.point.level" [9] "temperature:time.point" "temperature:time.point.level" [11] "model" "subCol" > fit1[6] $temperature.level [1] "20" "25" > fit1[8] $time.point.level [1] "20" "50" "90" > fit1[10] $`temperature:time.point.level` [1] "20:20" "20:50" "20:90" "25:20" "25:50" "25:90" [[alternative HTML version deleted]]