---
license: cc-by-4.0
pretty_name: CoRE MOF 2019
tags:
- chemistry
- materials-science
- metal-organic-frameworks
---

# CoRE MOF 2019

The public structures of the Computation-Ready Experimental MOF database (Chung et al., 2019),
version 1.1.4: metal-organic frameworks **determined by experiment**, cleaned to be used in
simulations. Their solvent is removed in two ways, which make the two splits:

| Split | Structures | Atoms | Atoms per structure |
|---|---|---|---|
| `asr`, all solvent removed | 12,020 | 3,926,702 | 10 to 10,560, median 192 |
| `fsr`, free solvent removed | 7,061 | 2,388,025 | 10 to 10,560, median 204 |

In `fsr` the solvent bound to the metals is kept. Most structures of `fsr` are also in `asr`,
with the same name up to its suffix.

## Labels

Structures come from experiment: the four labels (`total_energy`, `band_gap`, ...) are `NaN`.
The properties are the geometric ones of the table of the source, computed with Zeo++:

| `properties[...]` | Source column | Unit |
|---|---|---|
| `lcd`, `pld`, `lfpd` | largest cavity, pore limiting and largest free path diameters | Å |
| `density` | `cm3_g`, which holds the density | g/cm³ |
| `asa_m2_cm3`, `asa_m2_g` | accessible surface area | m²/cm³, m²/g |
| `nasa_m2_cm3`, `nasa_m2_g` | non accessible surface area | m²/cm³, m²/g |
| `void_fraction` | `AV_VF` | 0 to 1 |
| `av_cm3_g`, `nav_cm3_g` | accessible and non accessible volume | cm³/g |
| `has_oms` | 1 if the structure has an open metal site | 0 or 1 |
| `disorder` | 1 if the source put the structure in its `disorder` folder | 0 or 1 |

## Things to know

- **Very large cells.** One structure in ten has more than 690 atoms, 559 of `asr` have more
  than 1,000, and the largest has 10,560. The radius graph lists the pairs of atoms of a
  structure: at 5 Å a cell of 2,000 atoms takes 0.5 GB on a GPU, one of 5,000 atoms 3.3 GB.
  Filter by `num_atoms` or use batches of a few structures.
- **Nothing is filtered.** The 1,877 structures of `asr` and the 1,053 of `fsr` that the
  source set apart as disordered are kept, with `disorder` set to 1.
  `properties["disorder"] == 0` selects the others.
- **Experimental structures are not relaxed**, and hydrogens can be missing.
- **Atoms are in the order of the files.** 8 files of `asr` and 7 of `fsr` are not in P1:
  their symmetry operations were applied by pymatgen.
- **Names** are not in the file. `asr_index.parquet` and `fsr_index.parquet` give for each
  row the name of the structure in the source (a CSD refcode or the name of a file of
  supporting information, with a suffix), its metals and its DOI.
- No structure with partial occupancy was found.
- **Licence**: CC BY 4.0.

## Format

HDF5 files of flat concatenated arrays, the format of the other datasets of
[materials-toolkits](https://huggingface.co/materials-toolkits):

| Dataset | Shape | Type | Content |
|---|---|---|---|
| `lattice` | (n_struct, 3, 3) | float32 | lattice vectors as rows, in Å; the identity for a molecule |
| `num_atoms` | (n_struct,) | int64 | number of atoms of each structure |
| `ptr` | (n_struct,) | int64 | index of the first atom of each structure in `x` and `z` |
| `x` | (n_atoms, 3) | float32 | fractional coordinates; positions in Å for a molecule |
| `z` | (n_atoms,) | int64 | atomic numbers |
| `pbc` | (n_struct, 3) | bool | whether the structure is periodic along each lattice vector |
| `total_energy`, `formation_energy_per_atom`, `energy_above_hull`, `band_gap` | (n_struct,) | float32 | labels, NaN when unknown |
| `properties/<name>` | (n_struct,) | float32 | the properties listed above, NaN when unknown |

The position of an atom in Å is `x @ lattice`. The lattice of a crystal is rebuilt from the
lengths and the angles of its cell, in the standard orientation (first vector along x), and
its fractional coordinates are wrapped into the cell. The `batch` array is an artefact of the
writer and is not used when reading. Row *i* of an index file describes structure *i*.

## Usage

```python
from torchms.data import HDF5Dataset

dataset = HDF5Dataset.from_hub("materials-toolkits/core-mof", "asr.hdf5")
structures = dataset[:32]            # batched Structures
structures.properties                # the named properties, one value per structure
```

Built with [torchms](https://gitlab.univ-artois.fr/astrid.klipfel/torch-ms).

## Source and checks

- [CoRE MOF 2019 Dataset](https://doi.org/10.5281/zenodo.7691378), Zenodo, version 1.1.4:
  `2019-11-01-ASR-public_12020` and `2019-11-01-FSR-public_7061`, archives and tables, checked
  with their MD5.
- Built by `scripts/build_core_mof.py`. The atoms are read as they are listed in each file.
  Every structure was then compared with the reading of its file by pymatgen: cell, number of
  atoms, and each atom at the position of one atom of the same element. Three properties were
  compared with the table.
- Repository: [materials-toolkits/core-mof](https://huggingface.co/datasets/materials-toolkits/core-mof),
  files `asr.hdf5` and `fsr.hdf5`.