A deep-learning-based multiple toolkits (DeTool) approach that uses the inputs of enzymes and substrates for biocatalystic tasks.
This repository contains the PyTorch implementation of ALDELE framework.
ALDELE is a deep learning framework with two-phase attention and pairwise module to explicitly learn non-covalent local interactions between enzymes and substrates for biocatlytic purpose.
It works on two-dimensional (2D) molecular graphs and physicochemical properties of compounds, and target protein sequences with amino acid evolutionary matrix to perform prediction.
The source code developed in Python 3.10. The required python dependencies are given below.
numpy==1.23.1
packaging==21.3
pickleshare==0.7.5
py3Dmol==2.0.3
python-json-logger==2.0.7
rdkit==2023.3.2
scikit-learn==1.1.1
scipy==1.8.1
torch==1.12.0
torch-geometric==2.3.1
uri-template==1.3.0
urllib3==2.0.3
useful-rdkit-utils==0.2.7
Clone this Github repo and set up a new conda environment, or create a new project in pycharm with python 3.10.
# create a new conda environment
$ conda create --name ALDELE
$ conda activate ALDELE
# install requried python dependencies
$ pip install pandas
$ pip install numpy==1.23.1
$ pip install torch
$ pip install scikit-learn
# clone the source code of ALDELE and go into the model folder
The datasets folder contains all original experimental data used in ALDELE.
the code/Dataset_construction folder contains the guidance of data preproceesing procedure.
Please check the Data_construction_protocol.doc for details of each step.
To train ALDELE, we provide the dataset after basic submit bash file with hyperparameters in submit.sh of model/CPI_model.
The setting parameters in the submit.sh can be set as 1 to 6 for 6 different combinations of toolkits as we describe in the paper.
model6: full version, 2 sequence-based features + 2 ligand-based features + structure-based features
The pathdir need to modified based on the path you save the datasets. Simply using the following code to train the model:
$ bash submit.sh
The output results and model can be found in /output_modelx folders.
This implementation is inspired and partially based on earlier works [1] and [2].
[1] Tsubaki, Masashi, Kentaro Tomii, and Jun Sese. "Compound–protein interaction prediction with end-to-end learning of neural networks for graphs and sequences." Bioinformatics 35.2 (2019): 309-318.
[2] Li, Feiran, et al. "Deep learning-based k cat prediction enables improved enzyme-constrained model reconstruction." Nature Catalysis 5.8 (2022): 662-672.