matthodgman / drugstance

Calculate the semantic distance between drugs based on their indications using information theory.

Drugstance

Calculate the semantic distance between drugs based on the MeSH headings of their ChEMBL indications. We employ an information theory algorithm and the overlap coefficient. We decided to use an RBF kernel on the semantic distance and average it with the overlap coefficient to create our final semantic distance matrix.

Example usage (Slurm)

When running this pipeline on a large number of drugs, it is recommended to use SLURM to split computations across multiple jobs, each multiprocessing across 20 cores. To do this simple run the following command:

bash pipeline d2021.bin indications.tsv

Note: Files create and used in the pipeline will be written to a directory called data/ and output files will be written to a directory called output/.

Example usage (Docker)

This is a standalone version of the algorithm that runs within Docker. There is no multiprocessing nor job usage. Therefore it will run very slowly and require a lot of memory. We recommend running it with a smaller set of drugs and indications.

docker run --rm -v "$PWD":/data labsyspharm/drugstance:latest python3 /app/drugstance -i d2021.bin -m indications.tsv -o /data/

Input Files

d2021.bin is all MeSH data downloaded in ASCII format. indications.tsv is a TSV file from ChEMBL that contains in the column pref_name the name of the drug and in the column mesh_heading a valid MeSH heading that is an indication of that drug. For example:

pref_name	mesh_heading
TOFACITINIB	Immune System Diseases
TOFACITINIB	Arthritis, Rheumatoid
ASPIRIN	Pain

Transformations

Optionally, you can transform the output data using an RBF kernel (or implement your own transformation) and then take the average between all distance metrics to create a final semantic distance measurement between drugs.