OrbNet Denali: Molecular Geometries and DFT Energy Labels for ML Potentials

A training dataset for the OrbNet Denali machine learning potential, consisting of molecular geometries and corresponding energy labels. The data includes geometries in XYZ+ format and energy labels calculated at the wB97X-D3/def2-TZVP and GFN1-xTB levels of theory. It was created by researchers from Entos, Inc., Caltech, and NVIDIA for the 2021 paper "OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy".

Use Cases

• Training machine learning potentials for organic and biological molecules based on the provided geometry and energy data.
• Benchmarking quantum chemistry methods using the high-level DFT (wB97X-D3/def2-TZVP) reference energies.
• Studying molecular conformer energetics using the multiple geometry samples provided per molecule.
• Developing models that predict molecular properties from structure using the provided atomic coordinates and metadata.

Strengths

• Includes high-level Density Functional Theory (DFT) energy labels (wB97X-D3/def2-TZVP) as a reference.
• Provides multiple conformer geometries for many molecules, as indicated by the sample_id structure.
• Geometries include annotated molecular charge and spin multiplicity in the XYZ+ format.

Limitations

• Row count and total dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• All molecules are singlet states, limiting applicability to systems with different spin multiplicities.

Provenance

Source

Paper: "OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy" (2021).

Collection Method

Geometries and energy labels were calculated at DFT and semi-empirical levels of theory.