Location to share public data generated at Schrodinger, for validation of MLFF models.
All structural data is given in json to ease reading into python using the json module. Each test directory has multiple json files that collect structures that are related in some way and error statistics are typically computed over each file and then aggregated over the files. Each file is structured as a list of dicts and each dict corresponds to a single structure. Each dict has keys title, smiles, coordinates, elements and charge which give each structure a title, a smiles string, atomic coordinates in Angstrom, a list of elements corresponding to the coordinates and a net charge. In addition, if the structure is periodic there is a length 9 list lattice which gives the unit cell vectors in Angstrom. All structures have one or more energy keys, which are formated as E[<method>](<units>) where method is a string indicating the method and units is either Ha or Ry, for Hartree or Rydberg. In this data method is one of wB97X-D3BJ/def2-TZVPD, DLPNO-CCSD(T)/CBS or PBE-D3. Additional properties are given for some datasets and are detailed below where applicable.
This dataset contains 500 rotamer scans. Here we provide wB97X-D3BJ/def2-TZVPD and DLPNO-CCSD(T)/CBS estimates in Hartree. Please additionally cite the original work when using this dataset.
article{Rai2022TorsionNet
author = {Rai, Brajesh K. and Sresht, Vishnu and Yang, Qingyi and Unwalla, Ray and Tu, Meihua and Mathiowetz, Alan M. and Bakken, Gregory A.},
title = {TorsionNet: A Deep Neural Network to Rapidly Predict Small-Molecule Torsional Energy Profiles with the Accuracy of Quantum Mechanics},
journal = {Journal of Chemical Information and Modeling},
volume = {62},
number = {4},
pages = {785-800},
year = {2022},
doi = {10.1021/acs.jcim.1c01346},
note ={PMID: 35119861},
URL = {
https://doi.org/10.1021/acs.jcim.1c01346
},
eprint = {
https://doi.org/10.1021/acs.jcim.1c01346
}
}
This dataset contains 2109 rotamer scans of molecular fragments covering an expanded element set and includes some ionic examples. The reference energies are wB97X-D3BJ/def2-TZVPD and are given in Hartree.
This dataset contains a subset of the tautomer pairs in the tautobase set. The reference energies are wB97X-D3BJ/def2-TZVPD and are given in Hartree. Please additionally cite the original work when using this dataset.
article{Wahl2020Tautobase
author = {Wahl, Oya and Sander, Thomas},
title = {Tautobase: An Open Tautomer Database},
journal = {Journal of Chemical Information and Modeling},
volume = {60},
number = {3},
pages = {1085-1089},
year = {2020},
doi = {10.1021/acs.jcim.0c00035},
note ={PMID: 31967818},
URL = {
https://doi.org/10.1021/acs.jcim.0c00035
},
eprint = {
https://doi.org/10.1021/acs.jcim.0c00035
}
}
This dataset is a collection of loose PBE-D3 optimizations of 20 sets of organic crystals. The 20 systems were selected as challenging molecular systems from a prior study. Each set spans an energy range of up to 12 kcal/mol and reference PBE-D3 energies are given in Rydberg. Additionally, the number of molecules is given by the key number_of_molecules and it is customary to measure energy normalized to the number of molecules. For each set, any crystals matching experimental forms are indicated with the key experimental_match, whose value gives the name of the CCDC ref code, the RMSD N of this match is given by the key experimental_match_rmsd_n. When using this data please also cite the original work.
article{zhou2025CSP,
title={A robust crystal structure prediction method to support small molecule drug development with large scale validation and blind study.},
author={Zhou, Dong and Bier, Imanuel and Santra, Biswajit and Jacobson, Leif D. and Wu, Chuanjie and Garaizar Suarez, Adiran and Almaguer, Barbara Ramirez and Yu, Haoyu and Abel, Robert and Friesner, Richard A. and Wang, Lingle},
journal={Nat. Commun.},
volume={16},
number={2210},
year={2025},
doi={https://doi.org/10.1038/s41467-025-57479-1}
}
This dataset is a collection of tight PBE-D3 optimizations of 20 sets of experimentally observed organic crystals. The 20 systems were selected as challenging molecular systems from a prior study. Each set gives known Z'=1 crystal forms and reference PBE-D3 energies are given in Rydberg. Additionally, the number of molecules is given by the key number_of_molecules and it is customary to measure energy normalized to the number of molecules. When using this data please also cite the original work (given in the previous section).
this dataset gives equilibrated geometries of 62 examples of organic liquids. The systems were constructed to be about 1000 atom systems and average densities computed from MD simulations in the NPT ensemble should give some indication of the accuracy of a model to reproduce liquid densities. Each structure contains a property experimental_density(g/cc) and experimental_temperature(K), which give the mass density and temperature for which it was measured. These data were collected from the CRC handbook.
These datasets are described in the manuscript titled An Accurate Charge-Aware Machine-Learning Interatomic Potential for the Reduction of Li-ion Battery Electrolytes in Solution. Each contain 159 clusters of varying compositions of LixECy(PF6)z, geometries, and charge states (aside from dataset vi, which contains 258 separately sampled clusters). For a complete description see Figure 3 of the manuscript. Total energies, forces, and dipoles are provided at the wB97X-D3BJ/def2-TZVPD level of theory. If these data are used please cite the following manuscript:
@article{wei2025accurate,
title={An Accurate Charge-Aware Machine-Learning Interatomic Potential for the Reduction of Li-ion Battery Electrolytes in Solution},
author={Wei, Yujing and L. Weber, John and M. Stevenson, James and K. Goldsmith, Zachary and Xie, Xiaowei and D. Jacobson, Leif and A. Friesner, Richard},
year={2025},
publisher={ChemRxiv}
}
If using any of this data please cite our manuscript describing the data as well as MPNICE MLFF models.
@misc{Weber2025MPNICE,
title={Efficient Long-Range Machine Learning Force Fields for Liquid and Materials Properties},
author={John L. Weber and Rishabh D. Guha and Garvit Agarwal and Yujing Wei and Aidan A. Fike and Xiaowei Xie and James Stevenson and Karl Leswing and Mathew D. Halls and Robert Abel and Leif D. Jacobson},
year={2025},
eprint={2505.06462},
archivePrefix={arXiv},
primaryClass={physics.chem-ph},
url={https://arxiv.org/abs/2505.06462},
}