So will just list current design first and then get into the question:
rangedCounts <- dba.peakset(Adult_BN_count, bRetrieve=TRUE) nrows <- 1054182 ncols <- 12 counts <- matrix(runif(nrows * ncols, 1, 1e4), nrows) rowRanges<-GRanges(rangedCounts) sampleNames<-c("MBV1", "MBV2", "MBV3", "FBV1", "FBV2", "FBV3", "MBE7", "MBE8", "MBE9", "FBE1", "FBE2", "FBE3") sampleSex<-c("male", "male", "male", "female", "female", "female", "male", "male", "male", "female", "female", "female") sampleTreatment<-c("vehicle", "vehicle", "vehicle", "vehicle", "vehicle", "vehicle", "BB", "BB", "BB", "BB", "BB", "BB") colData<-data.frame(sampleName=sampleNames, sex=sampleSex, treatment=sampleTreatment) # Retrieve counts from dba.count without the site interval ranges counts <- as.matrix(mcols(rangedCounts)) # Construct a SummarizedExperiment se<-SummarizedExperiment(assays=list(counts=counts),rowRanges=rowRanges, colData=colData) # DESeq2 analysis table(sampleSex, sampleTreatment) dds<-DESeqDataSet(se, design= ~sex + treatment) dds$group <- factor(paste0(dds$sex, dds$treatment)) design(dds) <- ~0 + group dds <- dds[ rowSums(counts(dds)) > 1, ] dds <- DESeq(dds, betaPrior = FALSE)
So my question is about two types of results I am aiming for, group comparisons and pairwise comparisons, results method listed below:
#pairwise comparison BN_group_male<-results(dds, format = c("GRanges"), contrast=c("group", "maleBB", "malevehicle")) #group comparison resFemVeh_BN<-results(dds, lfcThreshold = 1, altHypothesis = "greater", format = c("GRanges"), contrast=c(-1/3, 1, -1/3, -1/3))
My overall goal here is to see:
1. Example: What are the male vs female diff irrespective of treatment etc.
2. Example: What are differential peaks specific to only the Male BB condition and no other.
I am wondering if this is the proper way to get the results for these particular comparisons given the above design?
Thanks!
My central question here is whether the design makes sense for the two ways in which I am retrieving results?
Also, in addition, I was noticing that when I lfcThreshold the #1 result type (a comparison of the BB vs vehicle in males) at the same cutoff as #2 I get pretty much no diff peaks (hence no lfc thresholding in the above #1 results formula), and wondering why that might be? Thanks again.
"Male vs female irrespective of treatment" you could get by averaging male vs female in each group. So putting 1/2 and -1/2 in the contrast. You don't seem to have anything like that in your code.
One vs the rest you can get with your second contrast.
Getting no diff peaks with cutoff: do some exploration and see if you find the answer. I would look at e.g. an MA plot of the results table.
Ok great thanks will put in that additional comparison, but in terms of the actual contrast I guess it would be the following, for male vs female only:
contrast=c(1/2, -1/3, -1/2, -1/3)
I guess just a little confused in this format for how to specify the comparison, not as much in the vignette.
Thanks again!
Put 1/2 for coefficients associated with M and -1/2 for the coefficients associated with F. This is a bit of a complex contrast because it isn’t represented by any coefficient in the design. There isn’t a M vs F “irrespective” when you are fitting a group for each combination of sex and treatment. You may want to discuss this in more detail with a statistical collaborator. I mostly provide software feedback here, but have limited time for further statistical instruction.
Thanks again!