lingluodlut / kesacnn

A domain Knowledge-enriched Self-Attention Convolutional Neural Network (KeSACNN) approach for PPI article classification.

KeSACNN

---

This repo contains the source code and dataset for the following papers:

• Ling Luo, Zhihao Yang, Lei Wang, Yin Zhang, Hongfei Lin and Jian Wang. KeSACNN: a protein-protein interaction article classification approach based on deep neural network. International Journal of Data Mining and Bioinformatics, 2019, 22(2): 131-148.
• Ling Luo, Zhihao Yang, Lei Wang, Yin Zhang, Hongfei Lin, Jian Wang, Liang Yang, Kan Xu and Yijia Zhang. Protein-Protein Interaction Article Classification: A Knowledge-enriched Self-Attention Convolutional Neural Network Approach. Procceding of 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2018.

Dependency package

KeSACNN uses the following dependencies:

• Python 2.7
• keras 2.0.3
• Tensorflow 1.1.0
• numpy 1.12.1

Content

• corpus

  • BioCreative II corpus
  • BioCreative III corpus

• src

  • attention_keras.py: the self-attention layer
  • BiLSTM.py: a BiLSTM baseline
  • CNN-Kim.py: a CNN baseline proposed by Kim
  • SACNN.py: Our Self-Attention CNN model
  • KeSACNN-arc1.py: Our KeSACNN-I model
  • KeSACNN-arc2.py: Our KeSACNN-2 model

Feature embeddings

The trained feature embeddings can be downloaded from https://www.kaggle.com/lingluodlut/kesacnn.

Train a basic SACNN model

To train a basic SACNN model, you need to provide the file of the training set, testing set and word embedding, and run the SACNN.py script:

python SACNN.py  

Train a KeSACNN-I model

To train our KeSACNN-I model, you need to provide the file of the training set, testing set, word embedding, BEN embedding and CUI embedding, and run the KeSACNN-arc1.py script:

python KeSACNN-arc1.py 

Train a KeSACNN-II model

To train our KeSACNN-II model, you need to provide the file of the training set, testing set, word embedding, BEN embedding and CUI embedding, and run the KeSACNN-arc2.py script:

python KeSACNN-arc2.py