Table 1. Network parameters used in our model.

A single-node network parameter provides two values to the feature vector per pair (8 single-node parameters create 16 values per pair). Each node-pair parameter contributes one value describing that pair. Parameter importance is measured using Gini importance[17,18] in the NetworkX Python package.[19]

Parameter	Context	Description	Parameter Importance
2nd degree shared neighbors	Single node	The sum of all nodes two edges away from the node of interest.	0.036, 0.030
Betweenness centrality	Single node	The sum of the fraction of shortest paths between two other nodes passing through the node of interest.	0.056, 0.056
Closeness centrality	Single node	The inverse sum of all shortest paths that originate at the node of interest.	0.035, 0.032
Communicability	Node pair	The sum of all closed walks between a pair of nodes.	0.043
Current-flow betweenness centrality	Single node	Analogous to betweenness centrality, but with all paths instead of shortest paths. Also known as random walk betweenness centrality.	0.057, 0.045
Degree centrality	Single node	The fraction of edges a node has of all possible edges.	0.074, 0.055
Eccentricity	Sindle node	The maximum distance from the node of interest to any other node in the network.	0.038, 0.035
Eigenvector centrality	Single node	The eigenvector for the largest eigenvalue of the matrix adjacency network.	0.042, 0.034
Inverse shortest path	Node pair	The inverse of the smallest number of edges connecting two nodes of interest.	0.048
PageRank	Single node	The rank of graph’s nodes based on the number of incoming links.	0.080, 0.072
Shared neighbors	Node pair	The intersection of two nodes’ sets of immediate neighbors.	0.067
Shared non-neighbors	Node pair	The number of nodes that are not immediate neighbors of both nodes of interest.	0.063