Description of noncovalent interactions from QTAIM-infused graph neural network

Data sets and trained models accompanying the paper "Description of noncovalent interactions from QTAIM-infused graph neural network". Code: https://github.com/ostreaterbl/NCI-QTAIM-GNN. data.zip: molecular graphs of dimers and complexes with the intermolecular bond paths and the QTAIM descriptors (ρ, ∇²ρ, G, V, H) at the bond critical points computed at the PBE0/def2-TZVP level, together with the reference interaction energies. It includes the training set (dimers from DES370K and IMP) and the external test sets S66, S66x8, BFDb (416 dimers), a DNA fragment and three neutral protein-ligand complexes from PLA15 (These large systems were calculated at the HF/def2-TZVP level). Each file is a Python pickle (a dictionary of PyTorch tensors); the format is described in the README of the code repository. MODELS.zip: checkpoints of the SchNet and PaiNN models for bond path classification, QTAIM descriptor regression and interaction energy prediction (global and bond-additive models). Each checkpoint stores the model weights and its full training configuration. Both archives follow the folder structure of the code repository and can be unpacked in its root. The geometries and reference energies are taken from the original data sets (DES370K, S66, S66x8, BFDb, PLA15); please cite the original publications when using them.