Datasets

ANAKIN-ME (Accurate NeurAl networK engINe for Molecular Energies)

Small organic molecules of carbon, hydrogen, nitrogen, and oxygen. Sampling was done via active learning. An ensemble of models flagged the conformations it disagreed on, and those were sent for DFT labelling.

size

~5.5M conformations of ~64k molecules

dft level

ωB97X/6-31G(d)

citation Smith et al., The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules (2020)

Open Catalyst 2020

Small adsorbates relaxing on catalyst surfaces drawn from 55 elements (mostly alloys and intermetallics).

size

~265M single points from ~1.3M relaxations

dft level

RPBE

citation Chanussot et al., Open Catalyst 2020 (OC20) Dataset and Community Challenges (2020)

Materials Project Trajectory (MPtrj)

Frames pulled from Materials Project relaxation and static calculations. A standard pretraining corpus for universal inorganic potentials.

size

~1.6M frames

dft level

PBE(+U)

citation Deng et al., CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling (2023)

Open Direct Air Capture 2023

Metal–organic frameworks with adsorbed CO₂ and H₂O. Built to train models for direct-air-capture sorbent discovery.

size

~39M single points from ~176k relaxations of ~8.4k MOFs

dft level

PBE-D3

citation Sriram et al., The Open DAC 2023 Dataset and Challenges for Sorbent Discovery in Direct Air Capture (2023)

Small-molecule/Protein Interaction Chemical Energies

Drug-like molecules from PubChem. Includes dipeptides, solvated amino acids, non-covalent dimers, and monatomic ion pairs. Assembled to train potentials for drug-like molecules interacting with proteins.

size

~1.1M conformations

dft level

ωB97M-D3(BJ)/def2-TZVPPD

citation Eastman et al., SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials (2023)

Open Materials 2024

Non-equilibrium inorganic structures. (Relaxed compositions rattled in position and cell, then sampled by short ab-initio molecular dynamics.) Built as large-scale pretraining data for interatomic potentials.

size

~110M structures

dft level

PBE(+U)

citation Barroso-Luque et al., Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models (2024)

Subsampled Alexandria

Relaxation frames of inorganic crystals from the Alexandria database, a machine-learning-guided search for new stable materials spanning most of the periodic table. Subsampled and released alongside OMat24 as fine-tuning data compatible with Materials Project DFT settings.

size

~11M frames

dft level

PBE(+U)

citation Barroso-Luque et al., Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models (2024)

Materials Potential Energy Surface

Structures sampled from MD snapshots of Materials Project compounds.

size

~505k structures (~435k in the PBE release)

dft level

PBE and r²SCAN

citation Kaplan et al., A Foundational Potential Energy Surface Dataset for Materials (2025)

Open Molecular Crystals 2025

Organic molecules packed into periodic cells of up to 300 atoms, spanning about a dozen elements. Aimed at polymorph ranking and crystal-structure prediction for pharmaceutical applications.

size

~27M structures (up to 300 atoms per cell)

dft level

PBE-D3

citation Gharakhanyan et al., Open Molecular Crystals 2025 (OMC25) Dataset and Models (2025)

Open Molecules 2025

Small molecules, biomolecules, electrolytes, and metal complexes computed at a high level of theory.

size

~100M single points across ~83M unique systems

dft level

ωB97M-V/def2-TZVPD

citation Levine et al., The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models (2025)