We are cocreating a "nodesphere" abstraction to pass around (get/put) subgraphs of mutual interest.
Nodesphere is intended especially for use by graph visualization tools like Core Network, Noomap, Metamaps...
I've already written prototypes of adaptors (client side coffeescript) which convert google spreadsheets and google drive folders into nodespheres (which are just graphs: nodes and edges).
Nodespheres will be content addressable and therefore automatically deduplicated. The data will likely be minified json with sorted keys.
Ultimately, I'd like to see these graphs stored and requested in post-server post-database systems like maidsafe.net and telehash.org . We can start with server/db prototypes tho.