Skip to content

Latest commit

 

History

History
65 lines (49 loc) · 2.65 KB

READEME.md

File metadata and controls

65 lines (49 loc) · 2.65 KB
Raw

TransPPI

TransPPI propose two types of deep transfer learning methods based on convolutional neural network (CNN) layers with pairwise position specific scoring matrix (PSSM) feature inputs to predict target human-virus protein-protein interactions (PPIs).

Usage

Step 0 Prepare PSSM files

cat ./x* > pssm.tar.gz tar -zxvf pssm.tar.gz

Step 1 Baseline method: pre-train DNN models (PSSM+CNN+MLP) on human-virus PPI source dataset.

./1_run_dnn.sh

Step 2 Two types of transfer learning methods and cross-viral tests.

a. Import and frozen the parameters of CNN layers of source domain, then train the DNN model (PSSM+CNN+MLP) on target dataset. ./2_run_transfer_frozen.sh

b. Import the parameters of CNN layers of source domain, then fine-tune parameters on target dataset. ./2_run_transfer_finetune.sh

c. Train the DNN model (PSSM+CNN+MLP) on a human-virus system and test the model on another human-virus system. ./2_run_cross_viruses.sh

Output

a. Run_result.txt - performance of various methods b. xx.h5 - deep learning models c. xx.txt - prediction result files containing label, pairwise protein ids and prediction score.

Notice

The human-virus systems and related parameters can be changed in the above shell scripts.

Dataset

Eight human-virus PPI systems. The ratio of positive-to-negative is 1:10. PSSM of human and viral proteins are deposited in 'script/pssm/'. 9880 human-HIV PPIs, 98800 non-human-HIV PPIs. 5966 human-Herpes PPIs, 59660 non-human-Herpes PPIs. 5099 human-Papilloma PPIs, 50990 non-human-Papilloma PPIs. 3044 human-Influenza PPIs, 30440 non-human-Influenza PPIs. 1300 human-Hepatitis PPIs, 13000 non-human-Hepatitis PPIs. 927 human-DENV PPIs, 9270 non-human-DENV PPIs. 709 human-ZIKV PPIs, 7090 non-human-ZIKV PPIs. 568 human-SARS-CoV-2 PPIs, 5680 non-human-SARS-CoV-2 PPIs.

Requirements

  • Tensorflow (==1.7.0)
  • Keras (==2.2.4)
  • scikit-learn (==0.22.1)
  • numpy (==1.16.6)

Use the following command to install all dependent packages

pip install requirements.txt

Reference

Part of the code was modified from https://github.com/muhaochen/seq_ppi and the corresponding reference is: Chen,M. et al. (2019) Multifaceted protein-protein interaction prediction based on Siamese residual RCNN. Bioinformatics, 35, i305–i314.

Citation

Please kindly cite the paper if you use refers to the paper, code or datasets. @article{Yang2021Transfer, title={Transfer learning via multi-scale convolutional neural layers for human-virus PPI prediction}, author={Yang, Xiaodi and Yang, Shiping and Lian, Xianyi and Wuchty, Stefan and Zhang, Ziding}, journal={xx} }