SchNet Model Embedding Vectors of QM9 Atoms Labelled According to Functional Groups Designation

(see README file for same description but better formatting) Embedding vectors (Embs0-Embs127) for all atoms in the first 10k molecules (indexed by mol_index) in the QM9 [1] dataset, generated by a trained SchNet model [2], for 6 successive different interaction layers. Embedding vectors are annotated by atom type (atomic_number), xyz coordinates (coordx,coordy,coordz), interaction layer in the network (layer), atomization energy (atomizationE) of the associated atom is given, and labelled according to functional group assignment (ldalabel). Also contains the model which the embedding vectors were extracted from (model1). The hyperparameters of the trained SchNet model are: 128 filters, 128 atom basis functions, 50 Gaussians, 6 interaction layers, and an interaction cutoff of 50 Angstroms. Model was trained on 100k training points (molecules) and 10k validation points of QM9. Best_model contains the trained model weights and parameters so that they can be loaded. Split.npz file can reproduce the training/validation data split, and lastly the traininglog.csv file contains information about training and validation loss over training epochs. [1] L. Ruddigkeit, R. van Deursen, L. C. Blum, J.-L. Reymond, Enumeration of 166 billion organic small molecules in the chemical universe database GDB-17, J. Chem. Inf. Model. 52, 2864–2875, 2012; https://doi.org/10.1021/ci300415d R. Ramakrishnan, P. O. Dral, M. Rupp, O. A. von Lilienfeld, Quantum chemistry structures and properties of 134 kilo molecules, Scientific Data 1, 140022, 2014. https://doi.org/10.1038/sdata.2014.22 [2] Schütt, K., Arbabzadah, F., Chmiela, S. et al. Quantum-chemical insights from deep tensor neural networks. Nat Commun 8, 13890 (2017). https://doi.org/10.1038/ncomms13890