SPICE 1.1.2: Quantum Mechanical Data for Drug-like Molecules and Peptides

SPICE is a collection of quantum mechanical data for training machine learning potential functions, with an emphasis on simulating drug-like small molecules interacting with proteins. It was created by researchers including Peter Eastman from Stanford University and described in a 2022 publication. The data includes molecular conformations, energies, gradients, and multipole moments for molecules identified by PubChem IDs, SMILES strings, or amino acid sequences.

Use Cases

• Training machine learning potential functions based on quantum mechanical energy and gradient data.
• Simulating interactions between drug-like molecules and proteins using the provided conformational and atomic property data.
• Analyzing molecular charge and multipole distributions using the MBIS charge, dipole, quadrupole, and octupole arrays.

Strengths

• Data is structured for machine learning with specific arrays for conformations, energies, and gradients.
• Includes multiple quantum mechanical properties per conformation, such as formation energy, DFT total energy, and molecular multipoles.
• Molecules are sourced from specific subsets like PubChem and include canonical SMILES strings with explicit hydrogens.

Limitations

• Row count and total dataset size are unknown, which may limit suitability assessment.
• Some data groups may be missing certain arrays if the MBIS calculation failed to converge.
• Last update date is unknown; freshness unverified.

Provenance

Source

Stanford University researchers, as described in the 2022 arXiv publication.

Collection Method

Quantum mechanical calculations, likely Density Functional Theory (DFT), for generated molecular conformations.