ANI-1

Overview

ANI-1 is a molecular reference dataset accompanied by a support repository containing scripts for accessing its records. The README cites a dataset publication describing 20 million calculated off-equilibrium conformations for organic molecules, alongside the original ANI-1 neural network potential paper. This entry concerns the dataset and its extraction utilities, not a runnable potential or a model-training framework.

The downloadable archive is linked through Figshare and expands into an ANI-1_release directory. Its data are distributed across eight HDF5 files named ani_gdb_s0x.h5, where x ranges from 1 to 8 and indicates the number of heavy atoms—C, N and O—in the molecules in that file. The documented quantities include molecular coordinates in Angstroms and energies in Hartrees. The README also supplies self-interaction atomic energy values for H, C, N and O.

The access workflow uses Python, NumPy and h5py. The supplied pyanitools.py includes an anidataloader class for loading and parsing the dataset, while example_data_sampler.py demonstrates sampling through that loader. The archive also includes classes described as supporting loading and storing data in the authors’ in-house format. These utilities provide a starting point for extracting reference data for downstream molecular modelling workflows.

The README specifies Python 3.5 or later and limits its stated reader testing to that version range; it does not establish compatibility with a particular current environment. The source excerpts do not document a complete record schema, training procedure or predictive performance evaluation. The repository’s MIT licence covers its software; dataset reuse terms are not established by these excerpts.

Key Features

• Eight HDF5 data files organize molecules by heavy-atom count, from one to eight C, N or O atoms.
• The `anidataloader` class in `pyanitools.py` loads and parses ANI-1 data.
• `example_data_sampler.py` demonstrates sampling records through the supplied loader.
• The archive includes Python classes for loading and storing data in the authors’ in-house format.
• The README specifies coordinate units in Angstroms, energy units in Hartrees and self-interaction atomic energy values for H, C, N and O.

Use Cases

• Suggested evaluation: extract coordinate–energy samples to assess ANI-1 as reference data for a molecular energy prediction workflow.
• Suggested evaluation: compare data subsets grouped by heavy-atom count when designing training and validation partitions.
• Suggested evaluation: inspect sampled off-equilibrium conformations for suitability in a study of molecular geometry–energy relationships.

How to Use

1. Read the support README to understand the archive structure, dependencies and units. Consult the cited dataset paper for scientific context and retain both requested citations when using ANI-1.
2. Obtain the dataset archive from the linked Figshare collection. Extract it using the README’s Unix-based archive procedure and locate the resulting ANI-1_release directory.
3. Prepare an environment with Python, NumPy and h5py. The README specifies Python 3.5 or later; check suitability for your own environment rather than treating that statement as current compatibility validation.
4. Follow the documented setup by adding ANI-1_release/readers/lib/ to PYTHONPATH. Inspect pyanitools.py and the example reader before running example_data_sampler.py as the documented access check.
5. Select the relevant ani_gdb_s0x.h5 files by heavy-atom count. Before downstream analysis, verify sampled records, preserve the documented units and check dataset reuse terms separately from the repository’s software licence.