Most of the interactions in GREG come from high-throughput technologies that can't offer directionality information by design (I think this is better represented by an undirected edge than by a double-arrow, and that is the way is treated in the biological literature). Biochemical directional information (arrows) usually design chains of reactions (so-called "pathways") and, in some cases, you may need the double-arrow to design pathways that go both ways. Such information is stored in a different type of databases, coming from different types of technologies (or predicted by algorithms).
If we decided to mix the two types of information in our database (interactions and pathways, undirected and directed), which we didn't, we may need to distinguish between directed, double-arrow, and undirected edges; and we may need the ability to have one undirected and one directed relationship between the same two nodes. An alternative would be to add pathway/direction information as properties, so the user could see that in some cases there is available information about direction, while in most cases not. That is just what I think now, but we haven't tried any of this yet.
Thanks for your comment.