Drug Repurposing Knowledge Graph (DRKG)

Drug Repurposing Knowledge Graph (DRKG) is a comprehensive biological knowledge graph relating genes, compounds, diseases, biological processes, side effects and symptoms. DRKG includes information from six existing databases including DrugBank, Hetionet, GNBR, String, IntAct and DGIdb, and data collected from recent publications particularly related to Covid19. It includes 97,238 entities belonging to 13 entity-types; and 5,874,261 triplets belonging to 107 edge-types. These 107 edge-types show a type of interaction between one of the 17 entity-type pairs (multiple types of interactions are possible between the same entity-pair), as depicted in the figure below. It also includes a bunch of notebooks about how to explore and analysis the DRKG using statistical methodologies or using machine learning methodologies such as knowledge graph embedding.

<img src="connectivity.png" alt="DRKG schema" width="600"> Figure: Interactions in the DRKG. The number next to an edge indicates the number of relation-types for that entity-pair in DRKG.

Statistics of DRKG

The type-wise distribution of the entities in DRKG and their original data-source(s) is shown in following table.

| Entity type | Drugbank | GNBR | Hetionet | STRING | IntAct | DGIdb | Bibliography | Total Entities | |:--------------------|---------:|------:|---------:|-------:|-------:|------:|-------------:|---------------:| | Anatomy | - | - | 400 | - | - | - | - | 400 | | Atc | 4,048 | - | - | - | - | - | - | 4,048 | | Biological Process | - | - | 11,381 | - | - | - | - | 11,381 | | Cellular Component | - | - | 1,391 | - | - | - | - | 1,391 | | Compound | 9,708 | 11,961 | 1,538 | - | 153 | 6,348 | 6,250 | 24,313 | | Disease | 1,182 | 4,746 | 257 | - | - | - | 33 | 5,103 | | Gene | 4,973 | 27,111 | 19,145 | 18,316 | 16,321 | 2,551 | 3,181 | 39,220 | | Molecular Function | - | - | 2,884 | - | - | - | - | 2,884 | | Pathway | - | - | 1,822 | - | - | - | - | 1,822 | | Pharmacologic Class | - | - | 345 | - | - | - | - | 345 | | Side Effect | - | - | 5,701 | - | - | - | - | 5,701 | | Symptom | - | - | 415 | - | - | - | - | 415 | | Tax | - | 215 | - | - | - | - | - | 215 | | Total | 19,911 | 44,033 | 45,279 | 18,316 | 16,474 | 8,899 | 9,464 | 97,238 |

The following table shows the number of triplets between different entity-type pairs in DRKG for DRKG and various datasources.

| Entity-type pair |:------------------- | (Gene, Gene) | (Compound, Gene) | (Disease, Gene) | (Atc, Compound) | (Compound, Compound) | (Compound, Disease) | (Gene, Tax) | (Biological Process, Gene) | (Disease, Symptom) | (Anatomy, Disease) | (Disease, Disease) | (Anatomy, Gene) | (Gene, Molecular Function) | (Compound, | (Cellular Component, Gene) | (Gene, Pathway) | (Compound, Side Effect) | Total | Drugbank | GNBR | Hetionet | STRING | IntAct | DGIdb | Bibliography | Total interactions | -------------------|-----------:|-------:|---------:|--------:|-------:|------:|-------------:|-------------------:| | - | 66,722 | 474,526 | 1,496,708 | 254,346 | - | 58,629 | 2,350,931 | | 24,801 | 80,803 | 51,429 | - | 1,805 | 26,290 | 25,666 | 210,794 | | - | 95,399 | 27,977 | - | - | - | 461 | 123,837 | | 15,750 | - | - | - | - | - | - | 15,750 | | 1,379,271 | - | 6,486 | - | - | - | - | 1,385,757 | | 4,968 | 77,782 | 1,145 | - | - | - | - | 83,895 | | - | 14,663 | - | - | - | - | - | 14,663 | | - | - | 559,504 | - | - | - | - | 559,504 | | - | - | 3,357 | - | - | - | - | 3,357 | | - | - | 3,602 | - | - | - | - | 3,602 | | - | - | 543 | - | - | - | - | 543 | | - | - | 726,495 | - | - | - | - | 726,495 | | - | - | 97,222 | - | - | - | - | 97,222 | Pharmacologic Class) | - | - | 1,029 | - | - | - | - | 1,029 | | - | - | 73,566 | - | - | - | - | 73,566 | | - | - | 84,372 | - | - | - | - | 84,372 | | - | - | 138,944 | - | - | - | - | 138,944 | | 1,424,790 | 335,369 | 2,250,197 | 1,496,708 | 256,151 | 26,290 | 84,756 | 5,874,261 |

Download DRKG

To analyze DRKG, you can directly download drkg by following commands:

wget https://dgl-data.s3-us-west-2.amazonaws.com/dataset/DRKG/drkg.tar.gz

If you use our notebooks provided in this repository, you don't need to download the file manually. The notebooks can automatically download the file for you.

When you untar drkg.tar.gz, you will see the following files:

./drkg.tsv
./entity2src.tsv
./relation_glossary.tsv
./embed
./embed/DRKG_TransE_l2_relation.npy
./embed/relations.tsv
./embed/entities.tsv
./embed/Readme.md
./embed/DRKG_TransE_l2_entity.npy
./embed/mol_contextpred.npy
./embed/mol_masking.npy
./embed/mol_infomax.npy
./embed/mol_edgepred.npy

DRKG dataset

The whole dataset contains four part:

drkg.tsv, a tsv file containing the original drkg in the format of (h, r, t) triplets.
embed, a folder containing the pretrained Knowledge Graph Embedding using the entire drkg.tsv as the training set and pretrained GNN-based molecule embeddings from molecule SMILES
entity2src.tsv, a file mapping entities in drkg to their original sources.
relation_glossary.tsv, a file containing rge glossary of the relations in DRKG, and other associated information with sources (if available).

Pretrained DRKG embedding

The DRKG mebedding is trained using TransE_l2 model with dimention size of 400, there are four files:

DRKG_TransE_l2_entity.npy, NumPy binary data, storing the entity embedding
DRKG_TransE_l2_relation.npy, NumPy binary data, storing the relation embedding
entities.tsv, mapping from entity_name to tentity_id.
relations.tsv, mapping from relation_name to relation_id

To use the pretrained embedding, one can use np.load to load the entity embeddings and relation embeddings separately:

import numpy as np
entity_emb = np.load('./embed/DRKG_TransE_l2_entity.npy')
rel_emb = np.load('./embed/DRKG_TransE_l2_relation.npy')

Pretrained Molecule Embedding

We also provide molecule embeddings for most small-molecule drugs in DrugBank using pre-trained GNNs. In particular, Strategies for Pre-training Graph Neural Networks develops multiple approaches for pre-training GNN-based molecular representations, combining supervised molecular property prediction with self-supervised learning approaches. We employ their method to compute four variants of molecule embeddings using DGL-LifeSci.

mol_contextpred.npy: From a model pre-trained to predict surrounding graph structures of molecular subgraphs
mol_infomax.npy: From a model pre-trained to maximize the mutual information between local node representations and a global graph representation
mol_edgepred.npy: From a model pre-trained to encourage nearby nodes to have similar representations and enforcing disparate notes to have distinct representations
mol_masking.npy: From a model pre-trained to predict randomly masked node and edge attributes

Tools to analyze DRKG

We analyze DRKG with some deep learning frameworks, including DGL (a framework for graph neural networks) and DGL-KE (a library for computing knowledge graph embeddings). Please follow the instructions below to install the deep learning frameworks.

Install PyTorch

Currently all notebooks use PyTorch as Deep Learning backend. For install other version of pytorch please goto Install PyTorch

sudo pip3 install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

Install DGL

Please install DGL (a framework for graph neural networks) with the following command. It installs DGL with CUDA support.

sudo pip3 install dgl-cu101

For installing other versions of DGL, please go to [Install DGL](https://docs.d

DRKG

Install / Use

README