Derify/msr-acc-tae25-regression

MSR-ACC TAE25 Regression Dataset

This dataset is a processed version of the Microsoft Research Accurate Chemistry Collection (MSR-ACC) TAE25 dataset, containing thermochemical data for molecules with computed Total Atomization Energies (TAE) using various quantum chemical methods.

Dataset Description

The original MSR-ACC TAE25 dataset contains QCSchema JSON files with 3D molecular geometries and computed thermochemical properties.

This processed version provides:

• SMILES representations generated from 3D coordinates and graph connectivity
• Train/validation/test splits for machine learning applications
  • Train set: Contains molecular structures from the original training partition (except test samples)
  • Validation set: Contains molecular structures from the original validation partition
  • Test set: 1,510 molecules sampled from the training set (1,485 random samples + 25 complete records)

Data Processing Pipeline

1. SMILES Generation

SMILES strings were generated from 3D molecular coordinates using the following process:

1. QCSchema Parsing: Loaded QCSchema JSON files using qcelemental for validation
2. Bonds (Graph Connectivity): Used pre-computed connectivity from graph:all fields in QCSchema extras
3. SMILES Generation: Generated canonical, isomeric SMILES using RDKit with MolToSmiles(canonical=True, isomericSmiles=True)

2. Dataset Splitting

The test set was created by sampling from the original training partition to ensure proper evaluation:

1. Sampling:

   • Selected 25 molecules with complete (non-NaN) target values across all thermochemical properties
   • Randomly sampled 1,485 additional molecules from remaining training data

2. Data Cleaning:

   • Maintained original train/validation split from source dataset

Data Format

Each Parquet file contains the following columns:

• filename: QCSchema JSON filename from the original dataset
• name: Molecular name from QCSchema
• smiles: Canonical, isomeric SMILES string
• symbols: Atomic symbols
• atomic_numbers: Atomic numbers
• geometry: 3D coordinates in Angstroms
• graph:all: Molecular connectivity information
• tae@*: Total Atomization Energy computed with various DFT methods and basis sets
• singlet-triplet-gap-*: Electronic excitation energies

Target Properties

The dataset includes Total Atomization Energies computed using:

• DFT Methods: B3LYP, B97M-V, M06-2X, r2SCAN, revDSD-PBEP86
• Basis Sets: def2-SVP, def2-TZVP, def2-QZVP, ma-def2-*
• Composite Methods: B97-3C, PBEh-3C, r2SCAN-3C
• Semi-empirical: GFN1-xTB, GFN2-xTB
• High-level: CCSD(T)/6-31G*, W1-F12

Usage Notes

• Missing Values: Many molecules have NaN values for specific methods (sparse coverage)
• SMILES Quality: Generated from optimized 3D geometries and graph connectivity
• Test Set: Contains both complete and incomplete records to enable various evaluation strategies

Citation

If you use this processed dataset, please cite both the original MSR-ACC publication and the Zenodo repository:

@misc{ehlert2025accuratechemistrycollectioncoupled,
      title={Accurate Chemistry Collection: Coupled cluster atomization energies for broad chemical space}, 
      author={Sebastian Ehlert and Jan Hermann and Thijs Vogels and Victor Garcia Satorras and Stephanie Lanius and Marwin Segler and Derk P. Kooi and Kenji Takeda and Chin-Wei Huang and Giulia Luise and Rianne van den Berg and Paola Gori-Giorgi and Amir Karton},
      year={2025},
      eprint={2506.14492},
      archivePrefix={arXiv},
      primaryClass={physics.chem-ph},
      url={https://arxiv.org/abs/2506.14492}, 
}

@dataset{ehlert_2025_15783826,
  author       = {Ehlert, Sebastian and
                  Hermann, Jan and
                  Vogels, Thijs and
                  Garcia Satorras, Victor and
                  Lanius, Stephanie and
                  Segler, Marwin and
                  Kooi, Derk P. and
                  Takeda, Kenji and
                  Huang, Chin-Wei and
                  Luise, Giulia and
                  van den Berg, Rianne and
                  Gori-Giorgi, Paola and
                  Karton, Amir},
  title        = {MSR-ACC/TAE25: 77k coupled cluster atomization
                   energies for broad chemical space
                  },
  month        = jul,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.15783826},
  url          = {https://doi.org/10.5281/zenodo.15783826},
}

License

This dataset is released under the Community Data License Agreement - Permissive - Version 2.0 (CDLA-Permissive-2.0), consistent with the original MSR-ACC dataset licensing.