I want to run DESeq2 to identify differences in open chromatin accessibility (i.e. ATAC-seq data) between 3 different GROUPS: A, B and C. I also wanted to control for SEX and BATCH effects.
I did the analysis two ways but obtained different results and would like to understand why this is the case, and what method is more appropriate. Note that I used the same ATAC-seq counts files (i.e. the total number of sites was the same) for both analysis methods.
Method 1: Combining all groups, using LRT while running DESeq2 to control for BATCH and SEX, and then obtaining pairwise results using contrast and the Wald test afterwards
Data<-DESeqDataSetFromMatrix(countData = ABCcounts, colData = ABCinfo, design = ~BATCH+SEX+GROUP)
Data2<-DESeq(Data, test = "LRT", reduced = ~BATCH+SEX)
AvsB<-results(Data2, independentFiltering = FALSE, test = "Wald", contrast=c("GROUP","A","B"))
AvsC<-results(Data2, independentFiltering = FALSE, test = "Wald", contrast=c("GROUP","A","C"))
BvsC<-results(Data2, independentFiltering = FALSE, test = "Wald", contrast=c("GROUP","B","C"))
Method 2: Running pairwise analyses independently. Note that the counts files are identical between Method 1 and Method 2 and I only deleted columns with counts for samples in the 3rd GROUP.
DataAB<-DESeqDataSetFromMatrix(countData = ABcounts, colData = ABinfo, design = ~BATCH+SEX+GROUP)
DataAB_2<-DESeq(DataAB, test = "LRT", reduced = ~BATCH+SEX)
AB<- results(DataAB_2, independentFiltering = FALSE)
DataAC<-DESeqDataSetFromMatrix(countData = ACcounts, colData = ACinfo, design = ~BATCH+SEX+GROUP)
DataAC_2<-DESeq(DataAC, test = "LRT", reduced = ~BATCH+SEX)
AC<- results(DataAC_2, independentFiltering = FALSE)
DataBC<-DESeqDataSetFromMatrix(countData = BCcounts, colData = BCinfo, design = ~BATCH+SEX+GROUP)
DataBC_2<-DESeq(DataBC, test = "LRT", reduced = ~BATCH+SEX)
BC<- results(DataBC_2, independentFiltering = FALSE)
Can someone please explain why the results between Methods 1 and 2 are not the same? Is the first method NOT taking into account BATCH+SEX? The main goal is to look for differences between GROUPS taking into account BATCH+SEX.
Thanks!
