solvechemistry/catecholbenchmark

---
license: mit
language:
- en
tags:
- che
size_categories:
- 1K<n<10K
---

# Summary

Catechol dataset for solvent selection and machine learning.

**NeurIPS Paper (Datasets and Benchmarks Track): [The Catechol Benchmark: Time-series Solvent Selection Data for Few-shot Machine Learning](https://arxiv.org/abs/2506.07619).**

# Data files

## Main data files

- **catechol_full_data_yields.csv**: Full data set with mixture solvents
- **catechol_single_solvent_yields.csv**: Only the single-solvent data
- **claisen_data_clean.csv**: Allyl Phenyl Ether data-set from an external source

## Lookup tables

Tables translating solvent names - as tabulated the main data files - to various pre-computed ML-readable representations:

- **acs_pca_descriptors_lookup.csv**: ACS Solvent Selection Guide's principle component analysis representation.
- **drfps_lookup.csv**: Fingerprint representation created using the difference in sets containing molecular substructures to the left and right of the reaction arrow in a SMILES string
- **fragprints_lookup.csv**: fragprints: A combination of molecular fingerprints, which are bit vectors indicating the presence of substructures in the molecule, and molecular fragments, which are count vectors indicating the number of times specific functional groups appear.
- **spange_descriptors_lookup.csv:** Representation based on measurable properties of solvents
- **smiles_lookup.csv**: SMILES strings.