Setting layers for a Dynamic Bayesian Network with bnstruct in R

I am currently creating a DBN using bnstruct package in R. I have 9 variables in each 6 time steps. I have biotic and abiotic variables. I want to prevent the biotic variables to be parents of the abiotic variables.For a Bayesian Network, it's pretty easy to implement using for instance layering = c(1,1,2,2,2) in learn.dynamic.network().

But a problem rises for the Dynamic part: I would like to keep preventing biotic variables to be parents of abiotic ones in every time step while preventing edges to appear between any variables from t+1 to t.

If I use in layering =:

1 for abiotic variables at t1
2 for biotic variables at t1
3 for abiotic variables at t2
4 for biotic variables at t2...

I allow biotic variables from t-1 to explain the abiotic variables at t (and I don't want that).

So I tried:

## 9 variables for 6 time steps 
test1 <- BNDataset(data = timedData,
                   discreteness = rep('d', 54),
                   variables = colnames(timedData),
                   node.sizes = rep(c(3,3,3,2,2,3,3,3,3), 6)
                   # num.time.steps = 6
                   )


## the 5 first variables are abiotic, the 4 last are biotics
dbn <- learn.dynamic.network(test1, 
                             num.time.steps = 6, 
                             layering = rep(c(1,1,1,1,1,2,2,2,2),6))

So now, I don't have any edges from biotic to abiotic (that's nice), but I have edges from variable_t(n+1) to variable_t(n).

I know that in bnlearn you can create a "blacklist" of edges that you don't want to see but I don't see any equivalent arguments in bnstruct. Any idea?

Solution

With the mmhc algorithm that is used as default, you can use the layer.struct parameter to specify which pairs of layers are allowed to have edges between them. layer.struct takes a binary matrix where cell i,j is 1 if there can be edges going from variables in layer i to variables in layer j, and 0 otherwise.

The best way to use this is to combine it with the manually-specified layering of your first solution.