wangjian-ucas/dltkcat

Deep learning based prediction of temperature dependent enzyme turnover rates

DLTKcat

DLTKcat v1.0: Deep learning based prediction of temperature dependent enzyme turnover rates

Dataset curation from SABIO-RK and BRENDA

The dataset curation process is in /code/GetData.ipynb.

How to use DLTKcat ?

1. Required inputs: substrate name, Uniprot ID of enzyme protein, temperature.
   
2. Get SMILES strings and enzyme protein sequences using convert_input(path, enz_col, sub_col ) in /code/feature_functions.py.
   
3. The input must be a csv file with columns of 'smiles', 'seq', 'Temp_K_norm', 'Inv_Temp_norm'.
   'Temp_K_norm' and 'Inv_Temp_norm' are normalized temperature and inverse temperature values.
   
4. Run prediction:
   

python predict.py --model_path [default = /data/performances/model_latentdim=40_outlayer=4_rmsetest=0.8854_rmsedev=0.908.pth]<br>
--param_dict_pkl [default = /data/hyparams/param_2.pkl] <br>
--input [input.csv] --output [output file name] <br>
--has_label [default = False]

5. Get attention weights of protein residues:
   

python get_attention.py --input [input.csv] --output [output file name]

Case studies

1. Mutants of Pyrococcus furiosus Ornithine Carbamoyltransferase via directed evolution (/data/PFOCT/,/code/CaseStudy_PFOCT.ipynb).
   Ref: https://doi.org/10.1128/jb.183.3.1101-1105.2001
2. Growth and metabolism of Lactococcus lactis and Streptococcus thermophilus at different temperatures(/data/GEMs, /code/GEMs.ipynb).
   Ref: https://doi.org/10.1038/srep14199, https://doi.org/10.1111/j.1365-2672.2004.02418.x

Dependencies

1. Pytorch: https://pytorch.org/
2. Scikit-learn: https://scikit-learn.org/
3. RDKit:https://www.rdkit.org/
4. BRENDApyrser: https://github.com/Robaina/BRENDApyrser
5. COBRApy: https://github.com/opencobra/cobrapy
6. Seaborn statistical data visualization:https://seaborn.pydata.org/index.html
7. Escher: https://github.com/zakandrewking/escher

Citation

DLTKcat: deep learning based prediction of temperature dependent enzyme turnover rates Sizhe Qiu, Simiao Zhao, Aidong Yang bioRxiv 2023.08.10.552798; doi: https://doi.org/10.1101/2023.08.10.552798

Issue

Users might encounter "Index out of range" error at amino_vector = self.embedding_layer_amino(amino).
The potential solution is +1 to n_atom, n_amino in model parameters, and train a new model.