DeepFE-PPI: An integration of deep learning with feature embedding for protein-protein interaction prediction we have reported a novel predictor DeepFE-PPI，a protein-protein interaction prediction method that integrates deep learning and feature embedding.For a given protein-protein pair, we first use Res2Vec to repressnt all residues, and then we regard the feature vector as the input of the network and apply a branch of dense layers to automatically learn diverse hierarchical features. Finally, we use a neural network with four hidden layers to connect their outputs for PPIs prediction.
===============================
DeepFE-PPI uses the following dependencies:
- Python 3.5.2
- Numpy 1.14.1
- Gensim 3.4.0
- HDF5 and h5py
- Pickle
- Scikit-learn 0.19
- Tensorflow 1.2.0
- keras 1.2.0
===============================
Datasets
The main directory contains the directories of S.cerevisiae, Human and five species-specific protein-protein interaction datasets. In each directory, there are:
- The S.cerevisiae core dataset: 5594 positive protein pairs and 5594 negative protein pairs.
- The human dataset: 3899 positive protein pairs and 4262 negative protein pairs.
- Each of five species-specific protein interaction datasets (C. elegans, E. coli, H. sapiens, M. musculus, and H. pylori) contains 4013, 6954, 1412, 313 and 1420 positive protein pairs.
===============================
model
The model directory contains 5 subfolders, with a folder for word2vec models and four folders for deep learning models, and structured as follows:
- c1c2c3: This folder contains a model corresponds to Section " Park and Marcotte’s evaluation scheme".
- dl: This folder have two subfolders:
- 11188 folder contains the model that the 5-flod cross validation methods executed on the S.cerevisiae core dataset.
- human folder contains the model that the 5-flod cross validation methods executed on the human dataset.
- rewrite: This folder contains a model corresponds to 'redo_cv_code.py'.
- train_11188_test_5_special: This folder contains a model corresponds to 'train_11188_test_5_special.py'.
- word2vec: This folder contains all word2vec models when Parameter Selection.
===============================
*.py
- 5cv_11188.py: 5-flod cross validation methods on the S.cerevisiae core dataset.
- 5cv_human.py: 5-flod cross validation methods on the human dataset.
- c1c2c3_11188.py: The code corresponds to Section " Park and Marcotte’s evaluation scheme". We redo the special multiple cross-validation method proposed by Park & Marcotte [Park Y, Marcotte EM. Nat Methods. 2012; 9 (12)]
- redo_cv_code.py: We rewrite the cross validation method without any libraries.
- swiss_Res2vec_val_11188.py: Parameter Selection for residue reprsetation and deep learning.
- train_11188_test_5_special.py: Codes that trains on the S.cerevisiae core dataset and tests on five species-specific protein interaction datasets.
===============================
Usage: Run these file from command line.
For example:
python train_11188_test_5_special.py output:
- accuracy_test_Celeg = 100,
- accuracy_test_Ecoli = 100,
- accuracy_test_Hpylo = 100,
- accuracy_test_Hsapi = 100,
- accuracy_test_Mmusc = 100
===============================
Contact us: Any questions about DeepFE-PPI, please email to dxqllp@163.com.
===============================
This dataset was used in the paper 'DeepFE-PPI: An integration of deep learning with feature embedding for protein-protein interaction prediction' for PPI prediction. For more details, please refer to the paper.