QM9S: 130K Organic Molecules with Calculated Spectra and Quantum Properties

130,000 organic molecules form the QM9Spectra dataset, derived from the popular QM9 benchmark. Author zihan zou and collaborators re-optimized molecular geometries and calculated a wide range of quantum mechanical properties, including scalars, vectors, tensors, and infrared, Raman, and UV-Vis spectra. Two data versions, .pt for PyTorch Geometric and .csv, are provided for training and use.

Use Cases

• Training graph neural networks for molecular property prediction based on calculated scalars like energy and NPA charges.
• Developing models for spectral generation or analysis based on provided infrared, Raman, and UV-Vis spectra.
• Benchmarking quantum chemistry methods using pre-calculated high-order tensors like polarizability and hyperpolarizability.
• Studying structure-property relationships in organic molecules using the re-optimized geometries and associated multi-fidelity property labels.

Strengths

• Contains 130,000 molecules, providing a substantial scale for machine learning tasks.
• Includes a diverse set of calculated properties, from scalars to 3rd order tensors, and three types of spectra.
• Offers data in both .csv and PyTorch Geometric (.pt) formats, catering to different user workflows.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for specific property subsets is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Derived from the QM9 dataset, with calculations performed by the author.

Collection Method

Molecular geometries were re-optimized and properties calculated using Gaussian16 at the B3LYP/def-TZVP level of theory.