# AIMNet2 (2025) Noncovalent Interaction Training Data (B97-3c), ColabFit Conversion

## Facts

- Provider: Nayal, Cho, Gao, Zheng and Isayev, Carnegie Mellon University; converted by ColabFit Exchange
- License: MIT (https://api.figshare.com/v2/articles/31141138)
- Size: 11 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/colabfit/AIMNet2
- Access: public, readable without an account
- Category: Chemistry & Materials
- Source: https://doi.org/10.26434/chemrxiv.15000203/v1
- Page: https://scigantic.com/archive/767c7ac4-93b2-5209-be76-612f88a8426c
- Keywords: Chemistry & Materials, DFT, interatomic potentials, Parquet, Hugging Face

## About

ColabFit Exchange conversion of the extended training dataset for AIMNet2 (2025), the second generation atoms-in-molecules neural network interatomic potential, released by Kamal Singh Nayal, Ilkwon Cho, Runtian Nick Gao, Peikun Zheng and Olexandr Isayev at Carnegie Mellon University (ChemRxiv preprint; data on CMU KiltHub). It targets noncovalent interactions (NCIs): hydrogen bonding, pi-pi stacking, dispersion, sigma-hole, ionic and electrostatic contacts. Structures come from SPICE v2.0.1 and CREMP macrocyclic peptides, PubChem molecules sampled by normal modes and metadynamics, and dimers built from Cambridge Structural Database monomers. Neutral and charged closed-shell systems contain up to 14 elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I) and up to 193 atoms. All calculations used ORCA 6.0.1 with the composite B97-3c DFT method (restricted Kohn-Sham, RIJCOSX). ColabFit lists 3,764,666 configurations with energy and atomic forces. Measured 2026-11-13: ds.parquet plus a co/ folder of 8 parquet shards (4.67 GB). Rows follow the ColabFit configuration schema (positions, atomic_numbers, cell, pbc, energy, atomic_forces, cauchy_stress, software, method and per-row metadata) and open with pandas.read_parquet, pyarrow or datasets.load_dataset; ds.parquet holds the dataset summary. Released under the MIT licence; credit Nayal et al. and the ColabFit Exchange.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("colabfit/AIMNet2")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `co/co_0.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/767c7ac4-93b2-5209-be76-612f88a8426c
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/767c7ac4-93b2-5209-be76-612f88a8426c
- Over MCP: `get_schema_card` with id `767c7ac4-93b2-5209-be76-612f88a8426c` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Chemistry & Materials datasets](https://scigantic.com/topics/chemistry-and-materials)
- [DFT datasets](https://scigantic.com/topics/dft)
- [Interatomic potentials datasets](https://scigantic.com/topics/interatomic-potentials)
