Prioritizing Repurposable Drugs for SARS-CoV-2 using Deep Learning and Population-based Validation

This is a code repository for project "Prioritizing Repurposable Drugs for SARS-CoV-2 using Deep Learning and Population-based Validation"

Requirements

Prepare necessary data

• Obtain Chemical-Gene Interactions, Genes, Phenotypes, and Pathways from COVID-19 curated list. Move the file under ./data/CTD/.

data/CTD/drug-gene-CTD_C000657245_ixns_20200703223915.tsv
data/CTD/pathways-CTD_C000657245_pathways_20200703225429.tsv
data/CTD/phenotype-drug-gene-CTD_C000657245_diseases_20200703224649.tsv

• Obtain virus-host protein-protein interaction from Gordon et al. Nature 2020

data/biology-database/baits-prey-mist.csv

• Download pre-trained DRKG embedding. Locate the files under ./data/DRKG/

data/DRKG/embed/DRKG_TransE_l2_entity.npy
data/DRKG/embed/entities.tsv
data/DRKG/embed/relations.tsv

Install packages

$ pip install torch-geometric
$ pip3 install torch

Notebooks for drug repurposing pipeline

We provided a self-contained notebook for easy dissemination

Preprocessing

We provided a step-by-step manual to preprocess data to node, edge, and node features.

Graph embedding and ranking model

• This notebook builds a graph (with Pytorch Geometric) and learns node representation using multi-relational variational graph autoencoders.
• After deriving the drug embedding, this notebook shows how to build a ranking model (Simple MLP + BPR loss), together with off-the-shelf baseline models.

Citation

Citation to be added