Datasets:
license: cc-by-nc-4.0
language:
- en
pretty_name: THEMol
size_categories:
- 1B<n<10B
tags:
- chemistry
- molecular-dynamics
- density-functional-theory
- medicine
- electrolytes
configs:
- config_name: hessian
data_files:
- split: full
path: Hessian/hessian_dataset.csv
- config_name: hessianrelax
data_files:
- split: full
path: HessianRelax/relax_dataset.csv
- config_name: torsionscan
data_files:
- split: full
path: TorsionScan/torsion_dataset.csv
- config_name: torsionscanrelax
data_files:
- split: full
path: TorsionScanRelax/torsion_relax_dataset.csv
- config_name: mbis
data_files:
- split: full
path: MBIS/mbis_dataset.csv
THEMol: Torsion, Hessian, Energy of Molecules
Dataset Summary
THEMol is an open-source collection of quantum mechanical properties tailored for organic molecules. It provides large-scale density functional theory (DFT) data for exploring intramolecular potential energy surfaces, including optimized geometries, structural relaxation trajectories, torsion scans, constrained torsion relaxation trajectories, Hessian matrices, and MBIS-derived atomic properties.
The dataset contains five tailored subsets and over three billion DFT calculations for molecules with up to 50 non-hydrogen atoms. Its chemical space spans twelve essential elements and diverse molecular architectures relevant to drug discovery, electrolytes, ionic liquids, and broader molecular modeling applications.
See the paper and GitHub repository for more details. The GitHub repository provides data readers, validation utilities, and statistical scripts.
Dataset Details
Dataset Description
- Repository https://github.com/ByteDance-Seed/THEMol
- Paper https://arxiv.org/abs/2605.14973
- Data format CSV index files plus HDF5 data files.
- Computation scale more than three billion DFT calculations across five subsets.
- Molecular scope organic molecules with up to 50 non-hydrogen atoms.
- Element coverage H, C, N, O, S, F, Cl, Br, Si, B, P, and I.
Subsets
| Subset | Level of Theory | Entries | Supplementary Metrics | Description |
|---|---|---|---|---|
| Hessian | B3LYP-D3(BJ)/DZVP | 3,102,537 | - | Optimized molecular geometries and corresponding Hessian matrices. |
| Hessian Relax | B3LYP-D3(BJ)/DZVP | 4,811,722 | 281,123,880 relaxation steps | Complete structural relaxation trajectories for the Hessian subset. |
| TorsionScan | B3LYP-D3(BJ)/DZVP | 4,192,791 | 2,436,985 molecules; 93,994,576 constraints | Comprehensive in-ring and non-ring torsional scans after constrained optimization. |
| TorsionScan Relax | B3LYP-D3(BJ)/DZVP | 4,914,677 | 3,090,560 molecules; 110,235,160 constraints; 2,993,685,868 steps | Complete constrained structural relaxation trajectories for the TorsionScan subset. |
| MBIS | PBE0/def2-TZVPD, or DZVP for I atoms | 3,082,151 | - | Atomic properties and model parameters from Minimal Basis Iterative Stockholder (MBIS) partitioning. |
Dataset Structure
The dataset is organized by subset. Each subset contains one CSV index file and HDF5 files referenced by the h5_file column.
/
├── Hessian/
│ ├── hessian_dataset.csv
│ └── *.h5
├── HessianRelax/
│ ├── relax_dataset.csv
│ └── *.h5
├── TorsionScan/
│ ├── torsion_dataset.csv
│ └── *.h5
├── TorsionScanRelax/
│ ├── torsion_relax_dataset.csv
│ └── *.h5
└── MBIS/
├── mbis_dataset.csv
└── *.h5
Data Fields
All HDF5 files are keyed by uuid. The CSV index files contain the UUID, mapped SMILES strings, and the HDF5 file location needed to retrieve each molecular record.
Hessian
CSV columns: uuid, mapped_nonisomeric_smiles, mapped_isomeric_smiles, h5_file
HDF5 structure:
/<uuid>/
mapped_nonisomeric_smiles utf-8 string
mapped_isomeric_smiles utf-8 string
atomic_numbers (N, 1) int32
coords (N, 3) float64
hessian (3N, 3N) float64
Hessian Relax
CSV columns: uuid, mapped_nonisomeric_smiles, mapped_isomeric_smiles, num_steps, h5_file
HDF5 structure:
/<uuid>/
mapped_nonisomeric_smiles utf-8 string
mapped_isomeric_smiles utf-8 string
atomic_numbers (N, 1) int32
step 0/
energy scalar float64
coords (N, 3) float64
forces (N, 3) float64
...
step k/
energy scalar float64
coords (N, 3) float64
forces (N, 3) float64
TorsionScan
CSV columns: uuid, mapped_nonisomeric_smiles, mapped_isomeric_smiles, torsion_indices, h5_file, num_constraints
HDF5 structure:
/<uuid>/
mapped_nonisomeric_smiles utf-8 string
mapped_isomeric_smiles utf-8 string
atomic_numbers (N, 1) int32
torsion_atom_indices (4,) int32
constraint 0/
energy scalar float64
coords (N, 3) float64
forces (N, 3) float64
constraint 1/
...
TorsionScan Relax
CSV columns: uuid, mapped_nonisomeric_smiles, mapped_isomeric_smiles, torsion_indices, h5_file, num_constraints, num_total_steps
HDF5 structure:
/<uuid>/
mapped_nonisomeric_smiles utf-8 string
mapped_isomeric_smiles utf-8 string
atomic_numbers (N, 1) int32
torsion_atom_indices (4,) int32
constraint 0/
energy (M,) float64
coords (M, N, 3) float64
forces (M, N, 3) float64
constraint 1/
...
Here M is the number of steps.
MBIS
CSV columns: uuid, mapped_nonisomeric_smiles, mapped_isomeric_smiles, h5_file
HDF5 structure:
/<uuid>/
mapped_nonisomeric_smiles utf-8 string
mapped_isomeric_smiles utf-8 string
atomic_numbers (N, 1) int32
coords (N, 3) float64
mbis_info/
atomic_volumes (N, 1) float64
atomic_charge (N, 1) float64
atomic_dipole (N, 3) float64
atomic_quadrupole (N, 3, 3) float64
parameters (M, 3) float64
Here M is the number of MBIS Slater functions.
For parameters, each row contains the 0-based parent atom index, the Slater-function charge population N_i, and the inverse width 1/sigma_i.
Units
| Quantity | Unit |
|---|---|
| Coordinates | Å |
| Energy | kcal mol^-1 |
| Force | kcal mol^-1 Å^-1 |
| Hessian | kcal mol^-1 Å^-2 |
| Charge | e |
| Dipole | e Å |
| Quadrupole | e Å^2 |
| Volume | Å^3 |
| 1/sigma_i | Å^-1 |
Licenses
Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).
Citation
If you use THEMol in your research or applications, please cite:
@misc{THEMol,
title={THEMol dataset: Torsion, Hessian, and Energy of Molecules},
author={Jiashu Liang and Tianze Zheng and Yu Xia and Xingyuan Xu and Xu Han and Zhi Wang and Siyuan Liu and Ailun Wang and Yu Liu and Shiqian Tan and Dongfei Liu and Zhichen Pu and Yuanheng Wang and Qiming Sun and Xiaojie Wu and Wen Yan},
year={2026},
eprint={2605.14973},
archivePrefix={arXiv},
primaryClass={physics.chem-ph},
url={https://arxiv.org/abs/2605.14973},
}