# Graph convolution with global information

Is it possible to add global information in addition to node information using Spektral or Pythorch?

For instance, information of nodes such as relative position and atomic mass give a molecular structure and its chemical property. In addition to nodes, I would like to add overall/global information that characterize the object... But, I do not find an example...

something like... (all nodes are fully connected)

NodeX ( global1, globa2, local1, loca2, local3 )
Node1 (val, val,   0,   0,  0 ) // global information
Node2 (0,     0, val, val, val) // local information
Node3 (0,     0, val, val, val) // local information
...
...


Does this make sense?

It seems Spektral and Pythorch are based on Message-passing neural networks. So, this does not accept global information?

I have to use Deep set or Full GN block?

If someone has good knowledge, I would like to hear it.