ProKcat

Multimodal Regression for Enzyme Turnover Rates Prediction. This paper has been published in [IJCAI 2025]. This is the code.

Data

BRENDA Release 2025.1 is now online. This new release includes: 168 new EC Classes and 1620 updated EC Classes 6,857 new primary literature references An updated metabolic pathway map featuring five new pathways: Glutathione-mediated detoxification, Curcuminoid biosynthesis, Monoterpenoid biosynthesis, Tropane alkaloid biosynthesis, Secologanin biosynthesis. Suggesting download the updated data in JSON and TXT formats [here].

File Specification

1. Data preparation: GetData.ipynb.from [DLTKcat].
2. Get SMILES strings and enzyme protein sequences features using code/gen_features.py, which will generate features using code/feature_functions.py.
3. run_esm_Kcat_finetune.ipynb and run_esm_Kcat_finetune.py is used to train and finetune esm. It may not be a good choice.
4. run_esm_Kcat.ipynb using protein esm embeddings and compound features, etc to predict Kcat.
5. run_train_test.ipynb, run KAN experiments on finetuned models.
6. Models in run_train_test.ipynb and run_esm_Kcat.ipynb can be merged to predict Kcat.

The KAN models or AI models' interpretability for science still have a long way to go.

Main Dependency

Pytorch (1.8.1+cu101)

Scikit-learn

esm

RDKit

BRENDApyrser

KAN

Thanks

Thanks for the work [DLTKcat]. The data and baseline models are mainly obtained from this repository.

Citation

If you find this project useful for your research, please use the following BibTeX entry.

@inproceedings{hu2025Multimodal,
  title={Multimodal Regression for Enzyme Turnover Rates Prediction.},
  author={Hu, Bozhen and Tan, Cheng and Li, Siyuan and Zheng, Jiangbin and Xia, Jun and Li, Stan Z.},
  booktitle={Thirty-fourth International Joint Conference on Artificial Intelligence (IJCAI 2025)},
  year={2025},
  organization={International Joint Conferences on Artificial Intelligence Organization}
}
@article{qiu2024dltkcat,
  title={DLTKcat: deep learning-based prediction of temperature-dependent enzyme turnover rates},
  author={Qiu, Sizhe and Zhao, Simiao and Yang, Aidong},
  journal={Briefings in Bioinformatics},
  volume={25},
  number={1},
  pages={bbad506},
  year={2024},
  publisher={Oxford University Press}
}