Installation:

These instructions will walk you through installing the program in the terminal.

• Requirements:
  • If you are using a Windows OS, you need to install and use WSL.
  • You need to have conda installed.

Clone the GitHub

git clone https://github.com/Collinformatics/CleaveNet

Create conda environment:

• If you are using MacOS run:

  conda env create -f environment_mac.yml
  

• If not, run:

   conda env create -f environment.yml
  

Activate the virtual environment:

  conda activate cleavenet

Test GPU activation:

  python testGPU.py

If you are using an NVIDIA GPU, you can monitor GPU usage with:

  watch -n 1 nvidia-smi

Note:

If the line "import cleavenet" gives you an error you'll need to add the working directory to PYTHONPATH:

export PYTHONPATH="$PYTHONPATH:$PWD"

All training data should be saved in the "data" directory.

Training:

Datasets should be placed in the "data" directory. Only the files names will need to be provided when training a model.

There are two available models, an LSTM and a Transformer. The Transformer is the default option. To use the LSTM add this flag:

--model-type lstm

Generator:

To train a model that can generate protein substrates:

python src/train_generator.py --data-path <filename>

Multiple parameters can be adjusted, to print the options run:

python src/train_generator.py --help

Predictor:

To train a model that can predict substrate activity:

python src/train_predictor.py --data-path <filename>

Multiple parameters can be adjusted, to print the options run:

python src/train_generator.py --help

• The external validation set is optional, if a file is not provided 5% of the dataset will be used for this set. To specify an external validation set use this flag:

  --data-pathEV <filenameEV>