# Catechol Benchmark Hackathon (NeurIPS 2025 DnB) — Data

来源：https://www.kaggle.com/competitions/catechol-benchmark-hackathon/data

采集时间：2026-10-20T03:20:02Z

Main data files:

We include two main data files for prediction. 

## catechol_full_data_yields.csv

This contains the full data set including mixtures of solvents. The columns are include:

- "EXP NUM": Number of experiment in which the data was collected
- "SOLVENT A NAME": Name of solvent in reservoir A
- "SOLVENT B NAME": Name of solvent in reservoir B
- "SolventB%": Percentage of solvent B used in the solvent mixture
- "Residence Time": Length of time the reaction was carried out in minutes
- "Temperature": Temperature at which the reaction was carried out in degrees celsius
- "SM": Yield of starting material measured, in % / 100
- "Product 2": Yield of product 2 measured, in % / 100
- "Product 3": Yield of product 3 measured, in % / 100
- "SM SMILES": SMILES string of starting material
- "Product 2 SMILES": SMILES string of product 2
- "Product 3 SMILES": SMILES string of product 3
- "SOLVENT A SMILES": SMILES string of Solvent A
- "SOLVENT B SMILES": SMILES string of Solvent B
- "SOLVENT A Ratio": Ratio of solvent in reservoir A if more than one was present e.g. if Solvent A = Water.Acetonitrile and "SOLVENT A Ratio" is [0.4, 0.6], this means reservoir A contained 40% water and 60% acetonitrile
- "SOLVENT B Ratio": Ratio of solvent in reservoir B if more than one was present e.g. if Solvent B = Water.Acetonitrile and "SOLVENT B Ratio" is [0.4, 0.6], this means reservoir B contained 40% water and 60% acetonitrile
- "Reaction SMILES A": Reaction SMILES under Solvent A
- "Reaction SMILES B": Reaction SMILES under Solvent B
- "RAMP NUM": Number of transient flow ramp at which the data was collected

## catechol_single_solvent_yields.csv: 

Only the single-solvent data with only a single solvent in the reaction. The columns are include:

- "EXP NUM": Number of experiment in which the data was collected
- "SOLVENT NAME": Name of solvent in reaction
- "Residence Time": Length of time the reaction was carried out in minutes
- "Temperature": Temperature at which the reaction was carried out in degrees celsius
- "SM": Yield of starting material measured, in % / 100
- "Product 2": Yield of product 2 measured, in % / 100
- "Product 3": Yield of product 3 measured, in % / 100
- "SM SMILES": SMILES string of starting material
- "Product 2 SMILES": SMILES string of product 2
- "Product 3 SMILES": SMILES string of product 3
- "SOLVENT SMILES": SMILES string of solvent used
- "SOLVENT Ratio": Ratio of solvent in the reservoir if more than one was present e.g. if Solvent = Water.Acetonitrile and "SOLVENT Ratio" is [0.4, 0.6], this means the solvent reservoir contained 40% water and 60% acetonitrile
- "Reaction SMILES A": Reaction SMILES

# Look-up tables

We also include a number of look-up tables in order to quickly find different parameterisations of the solvent molecules. These are:

* **acs_pca_descriptors_lookup.csv**: based on the PCA analysis from the ACS Green Chemistry Institute [Solvent Selection Tool](https://acsgcipr.org/tools/solvent-tool/).

* **drfps_catechol_lookup**: based on the differential reaction finger-print [DRFP](https://pubs.rsc.org/en/content/articlelanding/2022/dd/d1dd00006c) of the reactions under different solvents.

* **fragprints_lookup**: based on the concatenation of 'fragments' and 'finerprints' of molecules, as per [Griffiths et al.](https://pubs.rsc.org/en/content/articlelanding/2022/sc/d2sc04306h).

* **smiles_lookup.csv**: obtains the SMILES directly from the solvent names.

* **spange_descriptors_lookup**: based on the solvent properties as described in [Spange et al.](https://chemistry-europe.onlinelibrary.wiley.com/doi/full/10.1002/cmtd.202000039), imputing any missing values using a multi-task GP.

## Prediction goal

The goal of the competition is to achieve the best cross-validation score when training on the full data but leaving-out (a) full experiments in the case of mixture solvents, and (b) a single solvent out in the case of no mixture solvents. The models should accurately predict the values of "SM", "Product 2", and "Product 3" in the left out data.