In this work, we train small custom GPT on Moses and Guacamol dataset with next token prediction task. The model is then used for unconditional and conditional molecular generation. We compare our model with previous approaches on the Moses and Guacamol datasets. Saliency maps are obtained for interpretability using Ecco library.
- The processed Guacamol and MOSES datasets in csv format can be downloaded from this link:
https://drive.google.com/drive/folders/1LrtGru7Srj_62WMR4Zcfs7xJ3GZr9N4E?usp=sharing
- Original Guacamol dataset can be found here:
https://github.com/BenevolentAI/guacamol
- Original Moses dataset can be found here:
https://github.com/molecularsets/moses
- All trained weights can be found here:
https://www.kaggle.com/virajbagal/ligflow-final-weights
To train the model, make sure you have the datasets' csv file in the same directory as the code files.
./train_moses.sh
./train_guacamol.sh
./generate_guacamol_prop.sh
./generate_moses_prop_scaf.sh
If you find this work useful, please cite:
Bagal, Viraj; Aggarwal, Rishal; Vinod, P. K.; Priyakumar, U. Deva (2021): MolGPT: Molecular Generation using a Transformer-Decoder Model. ChemRxiv. Preprint. https://doi.org/10.26434/chemrxiv.14561901.v1
