QM7-X: Quantum-Mechanical Properties for Small Organic Molecules

QM7-X contains over 40 physicochemical properties for approximately 4.2 million molecular structures of small organic molecules with up to seven non-hydrogen atoms. The dataset includes both equilibrium and non-equilibrium structures, computed at the PBE0+MBD quantum-mechanical level of theory. It is designed to support the development of machine learning models for exploring chemical compound space and in silico molecular design.

Use Cases

• Training machine learning models for predicting atomization energies and dipole moments based on quantum-mechanical ground state quantities.
• Developing generative models for molecular design by learning from the exhaustive sampling of constitutional and stereoisomers.
• Benchmarking property prediction algorithms using tightly converged quantum-mechanical data for polarizability tensors and dispersion coefficients.
• Studying the relationship between molecular structure and physicochemical properties across a large region of chemical compound space.

Strengths

• Contains a large scale of ~4.2 million molecular structures, providing broad coverage of chemical compound space.
• Includes over 40 distinct physicochemical properties, ranging from global molecular to local atom-in-a-molecule quantities.
• Data is computed at a tightly converged, high-level quantum-mechanical theory (PBE0+MBD), ensuring accuracy.

Limitations

• The dataset is limited to small organic molecules with a maximum of seven non-hydrogen atoms (C, N, O, S, Cl), restricting its applicability to larger systems.
• Metadata regarding specific column names, exact file sizes, and precise row counts is not provided in the available sources.
• The last updated date is unknown, which may affect assessments of dataset currency.

Provenance

Source

University of Luxembourg (Author: Johannes Hoja)

Collection Method

Quantum-mechanical computations (PBE0+MBD level of theory) performed on systematically generated molecular structures.