This repository contains the source code of the three-stage approach for the chemical-protein interaction extraction task in the BioCreative challenge VI. Details of the three-stage approach are described in: Natural language processing based feature engineering for extracting chemical-protein interactions from literature, (2018), Lung P-Y, He Z, Zhao T & Zhang J.
- Python 3.4+
- Sklearn
- XGBoost
- NLTK
- Stanford Neural Network Dependency Parser
Partial dataset used in the model are located in the data folder for demonstration purpose. It contains abstracts of PubMed articles, tagged chemical/protein entities and labeled relations released by the task organizers. The complete dataset, as well as the gold standard for testing set, can be found at BiocreativeVI, or by contacting the organizers: Martin Krallinger & Jesús Santamaría.
In the last line of RunParser.py, specify the path to the Stanford Neural Network Dependency Parser. Next, run the command
$ sh demo.sh
This will run the pipeline, and generate ChemProtTest_sumbit.tsv, where each row contains: PubMedID, relation type, chemical entity, protein entity.