EpitopeVec: LinearEpitope Prediction Using DeepProtein Sequence Embeddings

Data, scripts and results for the EpitopeVec article by Bahai et al., Bioinformatics, 2021.

The epitope-prediction software is available at https://github.com/hzi-bifo/epitope-prediction

Requirements

• Python 3 with the following packages:

  • numpy 1.17.1
  • scipy 1.4.1
  • matplotlib 3.1.3
  • sklearn 0.22.1
  • pydpi 1.0
  • biopython 1.71.0
  • tqdm 4.15.0
  • gensim 3.8.3

  If these are not installed, you can install them with pip.

  pip3 install -r ./requirement/requirements.txt
  

  Additionally, pydpi 1.0 from pip might be incompatible with Python 3. Please install the pydpi package from the provided pydpi.tar.gz file.

  pip3 install pydpi.tar.gz
  

• Binary file for ProtVec representation of proteins can be downloaded using the following command in the protvec directory:

cd protvec
wget http://deepbio.info/embedding_repo/sp_sequences_4mers_vec.txt
wget http://deepbio.info/embedding_repo/sp_sequences_4mers_vec.txt.bin -O sp_sequences_4mers_vec.bin

Usage

• Clone this repository:

  git clone https://github.com/hzi-bifo/epitope-prediction-paper
  

• To train a new machine learning model, run the training file with name of the dataset you want to train on. The datasets are inside the retraining folder (bcpreds, ibce-el, lbtope and viral). eg:

   python3 retrain.py bcpreds
  

  Use bcpreds for training on the BCPreds dataset. Use lbtope for training on the LBTope dataset. Use ibce-el for training on the iBCE-EL training dataset.

Input

• For training a new model, two files containing a list of confirmed positive and negative epitopes are needed. These can be .txt files with each line containing a peptide. eg: In the ./retraining/bcpreds/ folder pos.txt contains a list of petides which are epitopes and neg.txt contains non-epitopes.
  For training domain-specific models, all the epitope petides should also be from the specific domain. eg: If one wants a viral-specific model, only include epitopes derived from viral proteins.

Output

• Training a new model will create a pickle file in the /retraining/input dataset folder. The modelname.pickle is the newly trained model which can be used with the EpitopeVec software (https://github.com/hzi-bifo/epitope-prediction) for testing.

Testing

• To test the performance of the trained models, please use the test.py file.

  python3 test.py model_pickle_file peptidefile
  
  

Here,

model_pickle_file is the .pickle file you trained previosly. peptidefile is a file with two columns. The first column is the peptide sequences and the second is the target (1 for epitope and 0 for non-epitope). See testing directory for a list of example files. The two columns should be tab-separated. Please provide the full name and location of the model_pickle file and the full name and location of the peptidefile. For example, if you want to test the model trained on the iBCE-EL datatset (svm-ibce-el.pcikle) on the ABCPred16 dataset run

python3 test.py ./retraining/ibce-el/svm-ibce-el.pickle ./testing/abcpred16.txt 

Results

The becnhmarking results of various methods are inside the results folder.