Using language models and ontology topology to perform semantic mapping of traits between biomedical datasets

Components

Set up BERT serving API

Set up main API

A complete setup of service components should look like (a user might need to change the various ports used by components)

• Main Vectology API: http://localhost:7560
• BERT model serving API: http://localhost:8560

Analysis

Setup

conda env create -f environment.yml
python -m spacy download en_core_web_lg
pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.4.0/en_core_sci_lg-0.4.0.tar.gz

UKB to EFO

Map UK Biobank traits to EFO using various models

python -m paper.ukb_efo

UKB to UKB

Pair-wise UK Biobank trait mappings using various models

python -m paper.ukb_ukb

BlueBERT-EFO analysis results

Set up a local development environment.

Then obtain the model package from repo releases and add to the dev environment (for details see docs above).

Then run the relevant analysis scripts (see the inference section of the docs above).