1. First create a virtual environment in your working directory called venv
2. Activate the virtual env
3. Install requirements with

pip3 install -r requirements.txt

4. Install apex as follows:

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir ./

5. Download the spacy english model

python3 -m spacy download en_core_web_sm

1. To create the training dataset by lexically annotating a corpus provided in pubtator format, run lexical_annot.sh

it requires the following arguments in order:

• The ontology in obo format
• The corpus that you want to annotate in pubtator format to create the weakly labeled dataset for training
• The output file name

Example:

./lexical_annot.sh ont.obo sample.input.txt weak_output.txt

2. Run the same command on a different corpus to create the development set
3. You will need to extract children and parents to use the ontology-based normalization, please refer to the following script for reference: https://github.com/smalghamdi/groovyScripts/blob/master/direct_parent_direct_child.groovy the output should be two files in json format
4. For ontology-based normalization you will also need to run the following script:

python3 get_allchildren.py direct_children.json

where the input is the direct_children json file

1. Run the train.sh script

it requires the output from the previous step which can be run twice: The first one to create training set and another one to create the development set

For the sake of the example we use the same file twice:

./train.sh weak_output.txt weak_output.txt 

Alternatively, you can use our sample files as follows:

./train.sh ../sample.input.4train.iob ../sample.dev.iob

1. Run the predict.sh script where the input is as follows in order:

• The pubtator file for which you want to predict
• The output file which will be stored under the output directory

To run the ontology-based normalization:

Replace the script name 'normalize.py' in the last line in predict.sh to 'normalize_ontology-based.py'

Example:

./predict.sh sample.input.txt final_output.txt

1. To use our trained models, download it from:

• Phenotype
• Disease

AND unzip it in the model/ directory

2. Download the intermediate files from:

• Phenotype
• Disease

AND unzip it in the ifiles/ directory