Skip to content

ChemistryLLMs/SMILES-probing

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Jun 13, 2024
1e9859c · Jun 13, 2024

History

11 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Chemical Language Models Have Problems with Chemistry: A Case Study on Molecule Captioning Task

📃 Paper

poster

Augmentations

'./code/augmentation.py' creates 4 types of augmentations:

  • rdkit canonicalization
  • explicit addition of hydrogens
  • kekulization
  • replacement of cycle identifiers by random numbers

Full description is provided in the paper.

Experimental dataset

Experimental dataset is provided in the folder "data" and was created by "augmentation" code. Original (non-augmented) sample of dataset is a test part of CHEBI-20.

Model evaluation

There are 4 models used in the experiment:

  • 'laituan245/molt5-base-smiles2caption'
  • 'laituan245/molt5-large-smiles2caption'
  • 'GT4SD/multitask-text-and-chemistry-t5-base-standard'
  • 'GT4SD/multitask-text-and-chemistry-t5-base-augm'

Code for model inference is located in the "code" folder.

References

If you use our repository, please cite the following related paper:

@inproceedings{probing,
  title={Chemical Language Models Have Problems with Chemistry: A Case Study on Molecule Captioning Task},
  author={Ganeeva, Veronika and Khrabrov, Kuzma and Kadurin, Artur and Savchenko, Andrey and Tutubalina, Elena},
  booktitle={The Second Tiny Papers Track at ICLR 2024},
  url={https://openreview.net/pdf?id=JoO6mtCLHD}
}