Required software

• Python v3.7.6 (Anaconda installation recommended)
• PyTorch v1.4.0
• scikit-learn v0.23.2
• Biopython v1.78
• RDKit 2020.09.1
• seaborn v0.11.0
• Matplotlib v3.3.2
• pandas v1.1.3
• SciPy v1.5.2
• NumPy v1.20.2

Usage

• For users who want to use the deep learning model for prediction, please run these command lines at the terminal:

  • (1). Download the DLKcat package

  git clone https://github.com/SysBioChalmers/DLKcat
  

  • (2). Download required Python package

  pip install numpy requests torch torchvision rdkit-pypi sklearn
  

  • (3). Change directory to DeeplearningApproach under the DLKcat package

  cd DLKcat/DeeplearningApproach
  

  • (4). Unzip the input.zip file under the Data directory

  unzip Data/input.zip
  

  • (5). Change directory to the Code/example under the DLKcat package

  cd Code/example
  

  • (6). Now you can use the trained deep learning model for your prediction via one command line. Here, one input file is needed to be prepared, please check the Code/example/input.tsv. For the input file, protein sequence should be provided, and users also need to provide substrate (compound) name or substrate (compound) SMILES, but substrate SMILES is recommended. If it is difficult to find the substrate SMILES, please provide the substrate name and leave the substrate SMILES blank

  python prediction_for_input.py input.tsv
  

  • Then the prediction results (output.tsv file) will be output under the Code/example directory

• For running analysis and regenerating all figures:

  • To regenerate all of the figures, unzip the input.zip file in Data/input.zip and run the corresponding figure functions in the Code/analysis directory

Preprocess

• For data collection and cleaning from the BRENDA database:
  • run the brenda_retrieve.py to get access to the web client and retrieve dataset from the BRENDA database
  • run the brenda_download.py to read all data in the retrieved files and output all EC files
  • run the findMaxKvalues_AllOrgs.py to read all EC files and find the max value for each substrate for the chosen microorganism
  • run the brenda_kcat_preprocess.py to generate Kcat data from all EC files into one file
  • run the brenda_kcat_clean.py to clean the dataset from the BRENDA database
  • run the brenda_sequence.py to get the protein sequence from BRENDA database by one example
  • run the brenda_sequence_organism.py to obtain the protein sequences for all data based on EC number and organism and output into one file for further use
  • run the brenda_get_smiles.py to get canonical SMILES just by substrate name for the BRENDA data using PubChem API
• For data collection and cleaning from the SABIO-RK database:
  • run the sabio_download.py to get access to the web client and download the dataset from the SABIO-RK database
  • run the sabio_kcat_unisubstrate.py to read all data from the downloaded files and output into one file for further use
  • run the sabio_kcat_clean_unisubstrate.py to clean the data by unifying all entries
  • run the sabio_kcat_clean.py to used to clean the data for the SABIO-RK data
  • run the sabio_kcat_unisubstrate_mutant.py to annotate the enzyme type information, i.e., wildtype or mutant
  • run the uniprot_sequence.py to to obtain protein sequence by uniprot protein id
  • run the sabio_get_smiles.py to get canonical SMILES just by substrate name for the SABIO-RK data and output one file for use
• For data combination based on the obtained dataset from the BRENDA and the SABIO-RK database:
  • run the combination_brenda_sabio.py to preliminarily combine the Kcat data from the BRENDA and the SABIO-RK database
  • run the combination_database_data.py to generate all the combined data into one file for deep learning and further analysis

Note

• For construction and evaluation of the deep learning model:
  • To see how the deep learning pipeline is constructed, check the corresponding functions in the Code/model directory
• For prediction of 343 yeast/fungi species via the deep learning model:
  • To obtain prediction results for 343 yeast/fungi species based on the trained deep learning model, unzip the input.zip file in Data/input.zip and run the corresponding function in the Code/prediction directory