Preclinical risk of bias assessment with CNN/Attention/HAN/BERT

Predict reporting scores and extract relevant sentences of five risk of bias items for preclinical publications:

• Random Allocation to Treatment/Control Group
• Blinded Assessment Outcome
• Conflict of Interest
• Compliance of Animal Welfare Regulations
• Animal Exclusions

Check the online demo

Usage

Clone source code

git clone https://github.com/qianyingw/pre-rob.git

Set environment

# Install miniconda3
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

# Create and activate virtual environment
cd pre-rob/rob-app
virtualenv -p python3 rob
source rob/bin/activate

# Install packages
pip install -r requirements.txt

# Download module & pre-trained weights
sh setup.sh

CSV file including txt paths as input

It should have two columns: 'id' and 'path'.

See input.csv for example. The 'path' column stores the absolute paths of TXT files.

python rob.py -p ../pre-rob/rob-app/example/input.csv  # absolutae path of input.csv
# Extract two relevant sentences for each item
python rob.py -p ../pre-rob/rob-app/example/input.csv -s 2  

Results are saved in output.csv.

Citation