OMC25

<h1 align="center" style="font-size: 36px;"> Open Molecular Crystals 2025 (OMC25)<br> Dataset </h1> <p align="center"> <img width="800" src="./omc25.svg"> </p> ## Dataset #### LICENSE: The OMC25 dataset is provided under a CC-BY-4.0 license OMC25 represents the largest high quality molecular crystal DFT dataset. OMC25 was generated at the PBE-D3 level of theory as implemented in Vienna Ab initio Simulation Package (VASP). OMC25 includes structures sampled from relaxation trajectories of molecular crystals generated by [Genarris 3.0](https://pubs.acs.org/doi/10.1021/acs.jctc.5c01080) starting from molecules in the [OE62 dataset](https://www.nature.com/articles/s41597-020-0385-y). For more details on the dataset, see the [paper](https://www.nature.com/articles/s41597-026-06628-2). Here we provide the OMC25 training and validation splits. Dataset files are written as ase-db [`LMDBDatabase`](https://gitlab.com/ase/ase-db-backends/-/blob/main/ase_db_backends/aselmdb.py) objects. | Splits | Size | Starting Molecular crystals | Starting Molecules | Storage | Download | | :-----: | :---: | :---: | :---: | :--: | :------: | | Train | 24,870,226 | 207,271 | 44,403 | 139GB | [train.tar.gz](https://dl.fbaipublicfiles.com/opencatalystproject/data/omc/250802/omc_train_250802.tar.gz) | | Val | 1,386,816 | 11,570 | 2,467 | 7.6GB | [val.tar.gz](https://dl.fbaipublicfiles.com/opencatalystproject/data/omc/250802/omc_val_250802.tar.gz) | We also provide further detailed information on all unique initial molecular crystal structures that underwent structural relaxations in [omc25-starting-crystals.csv](https://huggingface.co/datasets/facebook/OMC25-dataset/resolve/main/omc25-starting-crystals.csv) (description of its columns is shown below). | Column name | Description | |---|---| | `csd_refcode` | CSD reference code of molecule from OE62 dataset | | `z_value` | Number of molecular formula units in the crystal unit cell | | `genarris_step` | Sampled from generation (`gener`) or Rigid Press (`press`) step of Genarris 3.0 | | `xtal.id` | Unique crystal identifier among putative structures from Genarris step | | `split` | Structure was included in the training (`train`) or validation (`val`) split | | `nframes` | Number of frames sampled from relaxation trajectory | | `mol.composition`, `xtal.composition` | Composition of molecule and crystal, respectively | | `mol.natoms`, `xtal.natoms` | Number of atoms in molecule and crystal unit cell, respectively | | `mol.mass`, `xtal.mass` | Molar mass in g/mol of molecule and crystal unit cell, respectively | | `xtal.spacegroup` | Crystal space group with SYMPREC=10^-5 symmetry tolerance | ## How to read the data The OMC25 datasets can be accessed with the [fairchem](https://github.com/facebookresearch/fairchem) library. The package can be [installed](https://github.com/facebookresearch/fairchem#installation) with: ``` pip install fairchem-core ``` Once installed, a dataset can be read as follows ```python from fairchem.core.datasets import AseDBDataset dataset_path = "/path/to/omc25/dir/train" dataset = AseDBDataset({"src": dataset_path}) # index the dataset to get a structure atoms = dataset.get_atoms(0) atomic_positions = atoms.positions atomic_numbers = atoms.get_atomic_numbers() ``` Structures are stored as ASE [Atoms objects](https://wiki.fysik.dtu.dk/ase/ase/atoms.html). Each structure contains DFT total energy (eV), force (eV/Å), and stress (eV/ų) labels. Additionally, `atoms.info` contains several other properties and metadata that may be important for model development: `atoms.info:` ``` sid: XAZMUI-2-gener-b8a1abc1a6916c7-84 ## Unique structure identifier consisting of information below and also the index of the structure frame in the filtered relaxation trajectory csd_refcode: XAZMUI ## CSD reference code of molecule from OE62 dataset z_value: 2 ## Number of molecules in the crystal unit cell genarris_step: gener ## Sampled from generation (gener) or Rigid Press (press) step of Genarris 3.0 xtal.id: b8a1abc1a6916c7 ## Unique crystal identifier among putative structures from Genarris step source: omc25_data/dft_relax/batch1/XAZMUI/2/gener/b8a1abc1a6916c7 ## Unique internal identifier ``` ## Model checkpoints Baseline model checkpoints trained on the full OMC25 dataset can be found at [https://huggingface.co/facebook/OMC25](https://huggingface.co/facebook/OMC25) and [https://huggingface.co/facebook/UMA](https://huggingface.co/facebook/UMA). ## Support If you run into any issues feel free to post your questions or comments on [Github Issues](https://github.com/facebookresearch/fairchem/issues). ## Citation If you use this work, please cite: ``` @article{gharakhanyan2026open, title={Open Molecular Crystals 2025 (OMC25) dataset and models}, author={Gharakhanyan, Vahe and Barroso-Luque, Luis and Yang, Yi and Shuaibi, Muhammed and Michel, Kyle and Levine, Daniel S and Dzamba, Misko and Fu, Xiang and Gao, Meng and Liu, Xingyu and others}, journal={Scientific Data}, year={2026}, publisher={Nature Publishing Group UK London} } ```