The whole point of GNNs is that they generalize to arbitrary topologies by explicitly conditioning the idea of "neighbours" on the graph specifying the topology

The whole point of GNNs is that they generalize to arbitrary topologies by explicitly conditioning the idea of "neighbours" on the graph specifying the topology. Graph layout has been tried here https://github.com/limbo018/DREAMPlace https://github.com/limbo018/DREAMPlace to great fanfare although there is recent drama about it https://www.semanticscholar.org/paper/The-False-Dawn%3A-Reevaluating-Google's-Reinforcement-Markov/c43045140f4b30def64f04d1bc24a14030c0798b https://www.semanticscholar.org/paper/The-False-Dawn%3A-Reev... . Graph transformations are a thing as well https://arxiv.org/abs/2012.01470 https://arxiv.org/abs/2012.01470 but it's a tricky problem because you implicitly need to solve the graph matching problem