PubChem compounds before and after standardization

This repository contains a subset of 200k PubChem compounds before and after standardization.
This dataset is the basis of the PubChem-pretrained model presented in "Standardizing chemical compounds with language models" (available on ChemRxiv, see also the associated GitHub repository).

The associated pretrained model is also provided here, along with the splits used for training.

The data is provided under the CDLA-Sharing-1.0 license.

Provided files:

README.md: General README.

src_and_tgt_all.csv: All the compounds before and after standardization, in CSV format.

src-train.txt: The tokenized compounds before standardization in the train split.

tgt-train.txt: The tokenized compounds after standardization in the train split.

src-valid.txt: The tokenized compounds before standardization in the validation split.

tgt-valid.txt: The tokenized compounds after standardization in the validation split.

src-test.txt: The tokenized compounds before standardization in the test split.

tgt-test.txt: The tokenized compounds after standardization in the test split.

LICENSE.md: The details of the CDLA-Sharing-1.0 license.

pretrained_pubchem_step_120000.pt: the pretrained model.