Hello,

I have samples from 20+ patients from disease and normal tissues and would like to analyse the data normal vs disease with patient as a random effect. I have 2 questions:

1. Am I doing it correctly- below is my code
2. Due to poor QC some samples were removed, can I use incomplete pairs?

Here is my coldata (few rows)

Group Patient Age Sex Smoking Normal 1 57 Male Current Disease 1 57 Male Current Normal 2 64 Female Current Disease 2 64 Female Current Normal 3 65 Female Current Disease 3 65 Female Current

disease vs control

ddsMat <- DESeqDataSetFromMatrix(countData = count_matrix,
                                 colData = coldata,
                                 design = ~ Group)
dds <- DESeq(ddsMat)

adjusted for smoking

ddsMat <- DESeqDataSetFromMatrix(countData = count_matrix,
                                 colData = coldata,
                                 design = ~ Group+Smoking)
dds <- DESeq(ddsMat)
resultsNames(dds) 
[1] "Intercept"                 "Group_Normal_vs_Disease"             
[3] "Smoking_Former_vs_Current"

res <- results(dds, name="Group_Normal_vs_Disease")

disease vs normal - patient as a random effect

ddsMat <- DESeqDataSetFromMatrix(countData = count_matrix,
                                 colData = coldata,
                                 design = ~ Patient + Group)
dds <- DESeq(ddsMat)
resultsNames(dds) 

[1] "Intercept"           "Patient_10_vs_1" "Patient_11_vs_1" .....

 "Group_Normal_vs_Disease"
res  <- results(dds, name="Group_Normal_vs_Disease")#are these adjusted for patients ?

If you search around on the support site, you should be able to find previous answers addressing this topic in more depth. In short: DESeq2 can only use fixed effects and complete pairs, while limma's duplicateCorrelation() can accommodate random effects and incomplete pairs.