VGAE for node attribute generation

[ahariri13]

Thank you for sharing this interesting work. As most literature work on graph generative models focus on topology prediction (i.e P(A|Z), I was wondering how is it possible to make generative prediction on new node attributes as well (PA,X|Z). One example is the generative design of a molecular compound, where the decoder would output a molecular topology and its corresponding atomic (nodes) and bonds' (edges) attributes. In other words, I would be generating new compounds/ images/ whatever problem I am modeling rather than link predictions.

Your help is much appreciated,
Best regards