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

• Ling Luo, Zhihao Yang, Hongfei Lin, Jian Wang; Document triage for identifying protein–protein interactions affected by mutations: a neural network ensemble approach, Database, Volume 2018, 1 January 2018, bay097

BioCreativeVI-PM-Track Document Triage Task uses the following dependencies:

• Python 2.7
• keras 2.0.9
• numpy 1.12.1

• data
  • PPIm: the BioCreativeVI PM Track Document Traige corpus
  • PPI: the previous BioCreative PPI corpora
• src
  • Represent_luo.py
  • Eval.py
  • FileUtil.py
  • Load_dataset.py
  • PreProcessing.py
  • AttentionLayer.py
  • PPIAC-LSTM-pretrain.py: pre-train a PPI model
  • Hie_RNN.py: train a HieLSTM model
  • Hie_RNN-Classifier.py: classify the document using the HieLSTM model
  • PPIm-Feature-ppiac.py: train a PPIm model (including the LSTM, CNN, LSTM-CNN and RCNN models)
  • PPIm-Feature-ppiac-classifier.py: classify the document using the PPIm model (including the LSTM, CNN, LSTM-CNN and RCNN models)
  • Ensemble-LR.py: the ensemble using a LR model
  • Ensemble-voting.py: the ensemble using the voting
• fea_vocab
  • POS.vocab: the lookup table of the POS feature
  • NER.vocab: the lookup table of the NER feature

The trained models can be downloaded from https://www.kaggle.com/lingluodlut/biocreativevipmtrackmodels/data.

• models
  • BioCreativevi.rar: the 50-dimensional word embedding
  • bilstm-att-token-1l-50d-ppipre.rar: the pre-trained PPI model
  • models-ppipre-nofea.rar: the models without additional features (including LSTM, CNN, LSTM-CNN, RCNN, HieLSTM and a ensemble model using a logistic regression)