Hello,

I apologize that similar questions were asked many times before, yet I'm asking again. My design is a 3 x 2 design with a batch effect ( samples were not processed on the same day). I came across this post and am going to follow Dr. Smyth's suggestion (0 ~ + batch + group). I just wanted to know if my approach of making contrasts is correct.

My meta_data. I combined two factors (genotype and treat) into one factor, group.

         sample_name batch treat genotype    group
sample1      sample1    B1 untrt       WT WT_untrt
sample2      sample2    B1   trt       WT   WT_trt
sample3      sample3    B1 untrt       MR MR_untrt
sample4      sample4    B1   trt       MR   MR_trt
sample5      sample5    B1 untrt       GD GD_untrt
sample6      sample6    B1   trt       GD   GD_trt
sample7      sample7    B2 untrt       WT WT_untrt
sample8      sample8    B2   trt       WT   WT_trt
sample9      sample9    B2 untrt       MR MR_untrt
sample10    sample10    B2   trt       MR   MR_trt
sample11    sample11    B2 untrt       GD GD_untrt
sample12    sample12    B2   trt       GD   GD_trt

My design matrix

         batchB1 batchB2 groupGD_untrt groupMR_trt groupMR_untrt groupWT_trt groupWT_untrt
sample1        1       0             0           0             0           0             1
sample2        1       0             0           0             0           1             0
sample3        1       0             0           0             1           0             0
sample4        1       0             0           1             0           0             0
sample5        1       0             1           0             0           0             0
sample6        1       0             0           0             0           0             0
sample7        0       1             0           0             0           0             1
sample8        0       1             0           0             0           1             0
sample9        0       1             0           0             1           0             0
sample10       0       1             0           1             0           0             0
sample11       0       1             1           0             0           0             0
sample12       0       1             0           0             0           0             0

My contrasts

contrs <- makeContrasts(
  trt_vs_untrt_within_WT = groupWT_trt - groupWT_untrt,
  GD_vs_WT_within_untrt  = groupGD_untrt - groupWT_untrt,
  levels=colnames(design)
)

res_trt_within_WT <- glmQLFTest(fit, contrast=contrs[, "trt_vs_untrt_within_WT"])
res_GD_vs_WT_within_untrt = glmQLFTest(fit, contrast=contrs[, "GD_vs_WT_within_untrt"])

Q1. I'd like to know treatment effect within WT while accounting for the batch effect.

Is my res_trt_within_WT correct?

Q2. genotype effect (GD vs WT) within untreated

Is my res_GD_vs_WT_within_untrt correct?

I have gone through edgeR user guide as well as A guide to creating design matrices for gene expression experiments, yet I am asking these questions. I'm sorry...

unsolicited info:

I did PCA and observed the batch effect - PC1 separated batch 1 samples and batch 2 samples with 70% variance explained.

If you include 0+ in the model formula, then the order of the batch and treatment factors in the model becomes important. You need 0+group+batch rather than 0+batch+group.

See also https://www.biostars.org/p/9588610