ani1x / README.md
Zhongning's picture
Upload folder using huggingface_hub
a9ed09f verified
|
Raw
History Blame Contribute Delete
4.73 kB
metadata
license: cc0-1.0
tags:
  - ANI-1x
  - MACE
  - molecular dynamics
  - HDF5
  - extxyz
  - machine learning potential
language:
  - en
  - zh

ANI-1x

Dataset Description

The ANI-1x dataset is a molecular-configuration dataset for training and evaluation. The data package contains training, validation, and test HDF5 shards, along with a statistics JSON file and auxiliary extxyz files. Its fields are consistent with the energy and force fields in the MACE ANI-1x configuration, making it a standard benchmark dataset for learning and validating machine-learning interatomic potentials with MACE.

The element set consists of H, C, N, and O. The original ANI-1x data comes from ANI-1x datasets (FigShare); this repository is the companion data package for OneScience MACE training and evaluation examples.

The current data package contains approximately 30 files.

Supported Tasks

This standardized data repository provides the complete data package for the ANI-1x example, including training, validation, and test sets. It can be used for:

  • Training and validation of potential-energy-surface models for molecular configurations
  • Potential-energy-surface fitting for molecular dynamics simulations
  • Functional testing and benchmark evaluation of the MACE framework

Dataset Format and Structure

The data files are located in the data/ani1x/ directory:

File Path Format shape / Content Description
ani1x/ANI1x_cc_DFT_rc5_train/train_*.h5 HDF5 [N_frames, N_atoms] Training set, with 8 shards in total
ani1x/ANI1x_cc_DFT_rc5_val/val_*.h5 HDF5 [N_frames, N_atoms] Validation set, with 8 shards in total
ani1x/ANI1x_cc_DFT_rc5_test/Default__*.h5 HDF5 [N_frames, N_atoms] Test set, with 8 shards in total
ani1x/ANI1x_cc_DFT_rc5_statistics.json JSON Statistical information Contains atomic_energies, avg_num_neighbors, mean, std, atomic_numbers, and r_max
ani1x/ani1x_train.xyz extxyz [N_frames, N_atoms] Auxiliary training-set file
ani1x/ani1x_test.xyz extxyz [N_frames, N_atoms] Auxiliary test-set file
ani1x/ani1x_cc_dft.xyz extxyz [N_frames, N_atoms] Auxiliary file for the full dataset
metadata/sha256_manifest.txt SHA256 Checksum manifest File-integrity verification

HDF5 Data Format

The HDF5 file structure is config_batch_*/config_*. Each configuration contains:

Path dtype shape Description
atomic_numbers int64 [N_atoms] Atomic numbers
positions float64 [N_atoms, 3] Atomic coordinates, in Å
cell float64 [3, 3] Periodic cell matrix
pbc bool [3] Periodic boundary conditions
properties/energy float64 Scalar Total configuration energy, in eV
properties/forces float64 [N_atoms, 3] Atomic forces, in eV/Å

How to Use the Dataset

Download the dataset:

hf download --dataset OneScience-Sugon/ani1x --local-dir ./data

Official OneScience Information

Limitations and License

This dataset is the example data for MACE ANI-1x; it does not contain any newly generated samples or alter the original data. The original data comes from ANI-1x datasets (FigShare), with calculations performed at the wB97X/6-31G(d) level.

When using the ANI-1x dataset, please cite:

  • Smith, J. S.; Nebgen, B.; Lubbers, N.; Isayev, O.; Roitberg, A. E. Less Is More: Sampling Chemical Space with Active Learning. J. Chem. Phys. 2018, 148 (24), 241733. https://doi.org/10.1063/1.5023802
  • Smith, J. S.; Zubatyuk, R.; Nebgen, B.; Lubbers, N.; Barros, K.; Roitberg, A. E.; Isayev, O.; Tretiak, S. The ANI-1ccx and ANI-1x Data Sets, Coupled-Cluster and Density Functional Theory Properties for Molecules. Sci. Data 2020, 7, 134. https://doi.org/10.1038/s41597-020-0473-z