Can a hypernode be made of individual nodes, or conversely could a hyperedge be made up of individual hyperedges?

[syedzayyan]

Hi all,

Thanks for this amazing library firstly. Secondy I have a novice question. Could a hyperedge be potentially made up of smaller hyperedges? Take this example for example. A complex is a conjugate of multiple proteins, sometimes you have complexes of complex. The Protein is the individual normal node and then on top of there could a complex edge that is made of these protein nodes but sometimes complexes could be made of complexes. In those cases what could be the idea? Is this possible to encode in the library?

{'Complex2': {'Complex3': {'Protein1': 'features',
   'Complex4': {'Complex5': {'Protein2': 'features', 'Protein3': 'features'},
    'Protein4': 'features',
    'Protein5': 'features',
    'Protein6': 'features',
    'Protein7': 'features',
    'Protein8': 'features',
    'PhysicalEntity1': 'features'}},
  'Protein9': 'features'},
 'Complex6': {'Complex7': {'Protein10': 'features', 'Protein11': 'features'},
  'Complex8': {'Complex9': {'Protein12': 'features',
    'Protein13': 'features',
    'Protein14': 'features'},
   'Protein15': 'features'}}}

Kind Regards

[nwlandry]

Hi! Great question! This is a structure that some people call ubergraphs (https://arxiv.org/abs/1704.05547) and chygraphs (https://arxiv.org/abs/2209.07331). One could, in principle, do this programmatically (maybe as some kind of tree structure), but it would be hard to integrate into the XGI ecosystem. I definitely encourage you to play around with it, though!

[nwlandry]

Exactly. You sacrifice performance for generality, and there are also very few true algorithms for these types of networks. It is probably possible, but the efficient representation of these networks feels like the biggest obstacle. Maybe you could handle it by making everything an edge, and then having "up" and "down" relations for all of them, where "up" tells you the containing entity (like a hyperedge or complex) and "down" tells you what entities are contained (like nodes, etc.) For nodes, the down relation would be empty and for edges or complexes which aren't included in a larger entity, the up relation is empty.

[syedzayyan]

Would you be open to xgi having ubergraph/cygraph adaptation as maybe in seperate classes (seperate from the main xgi graph structure) as a contribution?

[nwlandry]

Sorry for the delay! Short answer is yes! I might recommend making a separate repository solely devoted ubergraph/cyrgraph, but using the XGI style to reduce the amount of overhead in this experimental phase. Then, once you've solidified more of the structure, it will become more clear whether it should be a standalone library or integrated into XGI. Thanks for the great idea!

[syedzayyan]

Hi @nwlandry! That sounds like a plan. I made a repo and am working on it 'slowly'. Thank you for all the help. I was wondering if there are any regular communication channels perhaps via email to keep in touch throughout the development, with you or anyone in your group working on something similar. If that is okay with you of course.

[nwlandry]

Awesome! Looks like a great start. We have a Zulip channel that you're welcome to join, though I must admit that I have been kinda quiet on there as of late 😅 the signup link is here