Relation Extraction from PubTator File

This program trains a model that identifies biological relations between entities using distant supervision. The pubtator annotated files are loaded from https://zenodo.org/record/1243969#.W7_xsRNKiu4 . The HIV-1 Human Interaction database is used for the distant supervision. https://www.ncbi.nlm.nih.gov/genome/viruses/retroviruses/hiv-1/interactions/

Getting Started

Virtual Environment is recommended to be built using requirements.txt

git clone (Link to this repository)
virtualenv <VE_PATH_NAME>
source <VE_PATH_NAME>/bin/activate
cd <PATH TO THIS REPOSITORY ON LOCAL MACHINE>
pip install -r requirements.txt

Prerequisites

Most important libraries
TensorFlow 1.5
NumPy 1.14

Running

DISTANT TRAINING

python relation_extraction_pubtator.py <MODE> <PUBTATOR_FILE> <DIRECTIONAL_DIRECTORY> <SYMMETRIC_DIRECTORY> <ENTITY_1_COLUMN> <ENTITY_2_COLUMN> <RELATION_COLUMN> <ENTITY_1> <ENTITY_2>

MODE: TRAIN_FEED_FORWARD TRAIN_RECURRENT

MODEL_DIRECTORY: Path to where to save model once trained

PUBTATOR_FILE: Pubtator annotated file in format from https://zenodo.org/record/1243969#.W7_xsRNKiu4

DIRECTIONAL_DIRECTORY: Directory with files for different directional relations (regulates,degrades,etc.) for distant supervision

SYMMETRIC_DIRECTORY: Directory with files for different symmetric relations (binds,colocalizes,etc.) for distant supervision

ENTITY_1_COLUMN: Column in distant supervision files (previous 2 commands) for Entity 1

ENTITY_2_COLUMN: Column in distant supervision files (previous 2 commands) for Entity 2

RELATION_COLUMN: Column in distant supervision files (previous 2 commands) for Relation of interest

ENTITY_1: one of the entities you're interested in, in format SPECIES_GENE i.e. 11676_GENE for HIV-1 genes (you can put NONE for species, NONE_GENE)

ENTITY_2: the other entities you're interested in, in format SPECIES_GENE i.e. 9606_GENE for HUMAN genes (you can put NONE for species, NONE_GENE)

PREDICTION

python relation_extraction_pubtator.py PREDICT <MODEL_DIRECTORY> <PUBTATOR_FILE> <ENTITY_1> <ENTITY_2> <RECURRENT_BOOL>

MODEL_DIRECTORY: Path of where model is trained__

PUBTATOR_FILE: Pubtator annotated file for predictions you want to make in format from https://zenodo.org/record/1243969#.W7_xsRNKiu4

ENTITY_1: one of the entities you're interested in, in format SPECIES_GENE i.e. 11676_GENE for HIV-1 genes (you can put NONE for species, NONE_GENE)

ENTITY_2: the other entities you're interested in, in format SPECIES_GENE i.e. 9606_GENE for HUMAN genes (you can put NONE for species, NONE_GENE)

RECURRENT_BOOL: Boolean if trained model you're predicting from is recurrent or feed forward. True for recurrent. False for Feed Forward