I have a following issue when running DESeq2 with a continuous effect variable.
In my experiment there are 2 time points and 2 groups. Each sample (individual) has a sample at timepoint 1 and timepoint 2, but belongs to only one of the groups.
dsgn <- "~1 + Batch + Group + Timepoint + Timepoint:Group + Timepoint:Group:X" dds <- DESeq2::DESeqDataSetFromMatrix( countData = countData, colData = colData, design = formula(dsgn) ) dds <- DESeq2::estimateSizeFactors(dds, "poscounts") dds <- DESeq2::DESeq(dds) resGAT0 <- results(dds, name = "GroupA.Timepoint0.X")
I am interested in the effect of a continuous variable X within each group A or B at each time point 0 or 1. I thought, the code above would achieve this. However, when inspecting results I found that many features (species) declared significant by DESeq2 (with a large logfold change) in the 'resGAT0' table had in fact all zero counts in the original data within that particular group A at time point 0. Is there some issue with the way I set up my DESeq2 analysis? In that case how to I obtain the actual effect for X within each of the 4 groups/time points combinations?
Is the issue related to the fact that the DESeq2::DESeq(dds) command returned a warning:
195 rows did not converge in beta, labelled in mcols(object)$betaConv. Use larger maxit argument with nbinomWaldTest