README.md

metadata

language:
  - en
task_categories:
  - text-generation
  - feature-extraction
tags:
  - chemistry
  - nmr
  - smiles
  - molecules
  - spectroscopy
pretty_name: SA-MolNMR Experimental Dataset
size_categories:
  - 1M<n<10M
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.parquet
      - split: validation
        path: data/validation.parquet
      - split: test
        path: data/test.parquet

SA-MolNMR Experimental Dataset

This dataset contains processed experimental molecular NMR records. Each example includes a SMILES string, molecular formula, atom count, and tokenized NMR input containing both proton NMR and carbon NMR data.

Dataset Details

The dataset is organized into train, validation, and test splits:

Split	Examples
Train	937,022
Validation	117,035
Test	117,205
Total	1,171,262

The train, validation, and test splits contain no overlapping id or smiles values.

Features

Each row contains the following fields:

Field	Type	Description
id	integer	Unique example identifier.
smiles	string	Molecular structure represented as a SMILES string.
atom_count	integer	Number of atoms in the molecule.
molecular_formula	string	Molecular formula.
tokenized_input	string	JSON string containing both 1HNMR and 13CNMR.
nmr_1h	string	JSON string containing the proton NMR peak list.
nmr_13c	string	JSON string containing the carbon NMR chemical shifts.

The 1HNMR field is represented as a list of peaks. Each peak follows this structure:

[chemical_shift, range_or_uncertainty, multiplicity, integration, coupling_constants]

The 13CNMR field is represented as a list of carbon chemical shifts.

Example

{
  "id": 503800,
  "smiles": "CN1C(=O)C2(C)Cc3ccccc3C2(CC(=O)c2ccc(Br)cc2)c2ccccc21",
  "atom_count": 52,
  "molecular_formula": "C26H22BrNO2",
  "tokenized_input": "{\"1HNMR\": [[7.635, 0.0, \"m\", \"3H\", []]], \"13CNMR\": [196.3, 173.3, 147.7]}",
  "nmr_1h": "[[7.635, 0.0, \"m\", \"3H\", []]]",
  "nmr_13c": "[196.3, 173.3, 147.7]"
}

Intended Uses

This dataset can be used for molecular representation learning, NMR-conditioned molecular modeling, structure elucidation, and related cheminformatics tasks.

Data Statistics

Across the dataset:

• tokenized_input is valid JSON for all examples.
• Every example contains both 1HNMR and 13CNMR.
• There are no missing values in the core fields.
• The average molecule contains about 42 atoms.
• The average example contains about 9 proton NMR peaks and 15 carbon NMR shifts.

License

Please specify the license before using or redistributing this dataset.