I have a network structure (adjacency matrix) and want to find a method that tells me which features of my network are most important to explain my response variable. In this case its an infection simulation measure. Intuitively, a neural network that can extract features of my graph that I can’t see and tells me its important would be great but I have no idea how to go about doing this. Ideally, I’d like to keep it simple too. Any advice is appreciated.