Hello,

I have an RNA-seq experiment whose colData looks like that (with more samples but this is a representation of all possible groups we have):

Sample State AAV Sex group

C001CH0 KI GFP M KI_GFP

C001CH2 WT GFP M WT_GFP

C001CH4 KI JAK F KI_JAK2

I would like to get the comparison of the three groups at a time with a one-way-anova approach from the column 'group'. Also, I am interested on studying these comparisons with the results function: list(c("AAV_JAK2_vs_GFP","StateKI.AAVJAK2"))) to get KI-GFP vs KI-JAK2, contrast=c("State","KI","WT")) to get GFP-WT vs GFP-KI and name="StateKI.AAVJAK2" to get WT-GFP vs KI-JAK2. The Sex column will be included in the formula to modulate the batch effect. Then, the design that I am considering for this is the following:

#1
dds_anova <- DESeqDataSetFromMatrix(countData = all_count,
                      colData = design_all,
                      design= ~ Sex + group)
dds = DESeq(dds_anova, test = "LRT", reduced = ~ 1) #to test the 3 groups all together
#This one apparently works

#2
dds_2by2 <- DESeqDataSetFromMatrix(countData = all_count,
                      colData = design_all,
                      design= ~ Sex + State + AAV + State:AAV) #to test the 2 by 2 after in the results

Error in checkFullRank(modelMatrix) : 
  the model matrix is not full rank, so the model cannot be fit as specified.
  Levels or combinations of levels without any samples have resulted in
  column(s) of zeros in the model matrix.

  Please read the vignette section 'Model matrix not full rank':

  vignette('DESeq2')

sessionInfo( )

R version 4.0.2 (2020-06-22)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: CentOS Linux 7 (Core)

DESeq2_1.28.1

I understand this is because deseq2 would expect also the WT-JAK2 existing when it creates automatically the possible contrasts. But that's not the case in my design and then, the matrix is not full of rank. I could split the matrix and the colData to have only the individual 2 by 2 comparisons individually. However, I don't know if the normalisation of the data in the full matrix containing all the information for the anova in comparison with the reduced version of the matrix could affect the statistics. Hence, I am not really being consistent with the normalisation to study the different situations.

• So: what do you think it could be the best approach to solve or hatch the problem? How could I avoid splitting the data to use all the information available in the matrix in order to normalise before obtaining the results for the different contrasts?

Thank you in advance,

Miriam

If it is not full rank then sex is at some point nested with group, meaning that for some group you have either only M or F in sex. Can you show the full colData (please highlight with markdown using the CODE button)?