That sounds reasonable to coding. But is it really meaningful for RDF ?
that is a question you have to answer yourself. the RDF is strictly speaking only defined for atoms in the first place. So you have to make some kind of approximation or additional definition to define what is the meaning of a molecular RDF. for a molecule like water, for example, just looking at the O-O g® is likely a very meaningful approximation. keep in mind, that g® is only looking at distances and assuming a spherical object and orientation invariant interactions.
otherwise, you would have to have a different distribution function, that includes more variables than just the distance between two objects.
again, if you want to reproduce what is published in a paper, you have to study what was done there.
axel.