I posted this on cross-validated but did not get a response. Trying my luck here. Sorry if this is not recommended.

I have an undirected graph with nodes separated within a specified distance, say d, being connected. The only feature of the nodes is an integer (0 or 1). The feature of an edge is the distance (as a vector, with 3 cartesian coordinates) between the nodes. My aim is to predict some new vector associated with ONLY the central node (all the nodes in the graph are with in a cutoff, say 5d, from the central node). Generally, this vector depends on the features of the nodes and edges in some unknown, nonlinear fashion.

enter image description here

This does not look like a graph-level prediction in GNN since the required vector is closely connected to a particular node in the graph structure. Can the problem be formulated in terms of node- or link-prediction? Please suggest if there is a way to study this problem using graph NN. Thanks in advance!

PS: Technically this is a physics problem, and I am interested in where the central atom will jump given the positions of all its neighbor atoms within a reasonable cutoff distance. The required vector depends on the type of atom (which is a feature of nodes) and its surroundings' atom type and position. I have a lot of data of central atoms with surrounding atoms (within cutoff distance) and the observed jumping vectors of central atoms, and want to build a ML model which predicts those jumping vectors given a central atoms and its surroundings. (this paragraph taken from the discussion of the same question I have posted elsewhere earlier).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.