metadata
license: cc-by-4.0
tags:
- chemistry
- molecules
- reactions
- smiles
- cheminformatics
size_categories:
- 1M<n<10M
task_categories:
- text-generation
- fill-mask
pretty_name: Reaction SMILES Dataset
dataset_info:
features:
- name: smiles
dtype: string
- name: source_id
dtype: int64
splits:
- name: train
num_bytes: 754819297
num_examples: 2779409
- name: validation
num_bytes: 94352378
num_examples: 347426
- name: test
num_bytes: 94352649
num_examples: 347427
download_size: 412910335
dataset_size: 943524324
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
Reaction SMILES Dataset
A collated dataset of 3.4M unique chemical reaction SMILES strings compiled from multiple public sources for use in pre-training and fine-tuning chemical language models.
Reaction SMILES (Simplified Molecular Input Line Entry System) extend the standard SMILES notation to represent complete chemical reactions. They encode reactants, reagents/catalysts, and products in a single text string using the > delimiter:
reactants>reagents>products
For example: CC(=O)O.CCO>[H+]>CC(=O)OCC.O represents the esterification of acetic acid with ethanol to form ethyl acetate.
Dataset Features
- smiles: The reaction SMILES string
- source_id: Original data source identifier
Data Sources
| Source ID | Source URL |
|---|---|
| 1 | US Patents 1976-Sep2016 Grants |
| 2 | US Patents 2001-Sep2016 Applications |
| 3 | CRD 1.37M Dataset (2024) |
| 4 | USPTO Year 2023 |
| 5 | Reaction SMILES Dataset (2023) |
License
This dataset aggregates publicly available reaction data. Please refer to the individual source links for specific licensing terms.