GNN Approximation of NP-Hard SPR Phylogenetic Tree Distances

A Graph Neural Network architecture approximates Subtree Prune and Regraft (SPR) distances in near-constant time. The model was trained on a dataset of 864 trees spanning up to 9,500 isolates to overcome the NP-hard computational complexity of exact distance calculations.