Hi all,

I am trying to figure out signals from DNA methylation for a continuous variable at multiple groups. We have mice samples at 4 time points in both sexes, and we are testing if DNA methylation positions respond to V1 using limma

I used the following design matrix and added duplicateCorrelation for random effect. I am asking the following three questions:

1. which DNA methylation positions respond to V1 in females?
2. which DNA methylation positions respond to V1 in males?
3. which DNA methylation positions respond to V1 differently between females and males?

V1 = meta_dat$variable
sex = meta_dat$sex
X = ns(meta_dat$time_point, df = 3)
design = model.matrix(~ 0 + sex + sex:V1 + X)

Q1: Does this design matrix make any sense?

The results from limma gave the following columns

colnames(coef(fit2))
[1] "sexfemale"       "sexmale"         "X1"              "X2"              "X3"             
[6] "sexfemale.V1" "sexmale.V1"

To answer the above questions,

1. topTable(fit2, coef = 6, n = Inf, adjust.method = "BH")
   

2. topTable(fit2, coef = 7, n = Inf, adjust.method = "BH")
   

3. cont_sexdiff = makeContrasts(sexfemale.V1 - sexmale.V1,
                         levels = colnames(coef(fit2)))
   fit_cont = contrasts.fit(fit2, cont_sexdiff)
   fit_cont = eBayes(fit_cont)
   topTable(fit_cont, n = Inf, adjust.method = "BH")
   

Q2: Is my approach appropriate? Thank you.

Yes, the design and contrasts are correct.

Your analysis is also adjusted for covariates defined by X, which I assume is what you want to do.