Yes, this is possible, and AFAIK is possible in DESeq2, too. It's easy to get tripped up by R builds design matrices from the symbolic formula that contain interactions. When in doubt, expand the interactions explicitly with e.g. `interaction`

.

```
library(MAST)
ng = 10
nc = 20
cd = expand.grid(A = factor(1:3), B = factor(1:5), rep = 1:nc)
cd$AB = with(cd, interaction(A, B))
mm = model.matrix(~AB, data = cd)
eta = mm %*% c(3, -2, 0, # (A=2, B=1) - (A=1, B=1), not tested
2, #(A=1, B=2) - (A=1, B=1), tested
rep(0, 11)) # rest null
U = matrix(eta, nrow = length(eta), ncol = ng)
U = U + rnorm(length(U))
PV = as.vector( exp(eta) / (1 + exp(eta)))
V = 1*(runif(length(U)) < PV)
dim(V) = dim(U)
Y = U*V
sca = FromMatrix(t(Y),cData = cd)
zz = zlm(~AB, sca = sca)
waldTest(zz, Hypothesis("AB2.5 - AB1.2"))
# Similar to (up to sampling error)
waldTest(zz, Hypothesis("-AB1.2"))
# The AB2.5 coefficient should be approx. zero
waldTest(zz, Hypothesis("AB2.5"))
```

Really, this is more of a base-R question. Forsake not your prophets Venables and Ripley. Chapter 6 shall provide the Light. the Truth, and the Way.