AIMNet2(2025) is the extended training dataset for the AIMNet2 (second generation atoms-in-molecules network) neural network interatomic potential, curated to improve the model's description of noncovalent interactions (NCIs) including hydrogen bonding, pi-pi stacking, dispersion, sigma-hole, ionic, and electrostatic contacts. The dataset covers neutral and charged closed-shell molecular systems composed of up to 14 non-metal elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I) with up to 193 atoms per system. Structures were drawn from three complementary sources: (a) molecular geometries from SPICE v2.0.1 (solvated systems, amino acid-ligand pairs, water clusters) and the CREMP dataset (macrocyclic peptides); (b) small neutral and charged molecules from PubChem sampled via normal mode sampling and metadynamics-guided geometry exploration; (c) dimer geometries assembled from Cambridge Structural Database (CSD) monomers (up to 14 supported elements, fewer than 200 atoms) and pre-optimized with AIMNet2-wB97M-D3(2023) to remove steric clashes while preserving configurational diversity. All quantum chemical calculations used ORCA 6.0.1 with the composite B97-3c DFT functional under restricted Kohn-Sham (RKS) formalism. SCF convergence was enforced with TightSCF and SlowConv; RIJCOSX integral acceleration and DEFGRID2 integration grid were applied throughout. AIMNet2(2025) was initialized from AIMNet2(2023) weights and continually pretrained on this dataset without weight freezing or regularization, using a multi-task loss over energy (w=1.0), forces (w=0.2), and Hirshfeld partial charges (w=0.5).

Dataset authors

Kamal Singh Nayal, Ilkwon Cho, Runtian Nick Gao, Peikun Zheng, Olexandr Isayev

Publication

https://doi.org/10.26434/chemrxiv.15000203/v1

Original data link

https://doi.org/10.1184/R1/31141138

License

MIT

Number of unique molecular configurations

3764666

Number of atoms

130288462

Elements included

As, B, Br, C, Cl, F, H, I, N, O, P, S, Se, Si

Properties included

energy, atomic forces

Usage

ds.parquet : Aggregated dataset information.
co/ directory: Configuration rows each include a structure, calculated properties, and metadata.
cs/ directory : Configuration sets are subsets of configurations grouped by some common characteristic. If cs/ does not exist, no configurations sets have been defined for this dataset.
cs_co_map/ directory : The mapping of configurations to configuration sets (if defined).

ColabFit Exchange documentation includes descriptions of content and example code for parsing parquet files:

Downloads last month: -

Total file size:

4.67 GB

Cite this dataset Nayal, K. S., Cho, I., Gao, R. N., Zheng, P., and Isayev, O. AIMNet2. ColabFit, 2026. https://doi.org/None

This dataset has been curated and formatted for the ColabFit Exchange

This dataset is also available on the ColabFit Exchange:

Visit the ColabFit Exchange to search additional datasets by author, description, element content and more.

Dataset Name

Description