COVID19 Open Targets data integration

February 2022 Update:

The Open Targets COVID-19 Target Prioritisation Tool has been deprecated and archived. For more information, please see opentargets/platform#1964 and this Community thread.

Aim

Centralise publicly available datasets in order to build a Virus – Host Target Knowledgebase for Drug Target Selection.

A full description of the project vision is here Some of the questions the project is trying to answer here

Contributors

This project is now closed to contributors.

Data flow

1. Potential data sources are fetched based on URLs provided in the Makefile if such URL is not available, the data tables can be directly added to the /data folder.
2. Relevant pieces of information is extracted from the raw source data as part of a parsing step.
3. Data might be further processed if necessary eg. mapping cross-references, integration with other sources etc.
4. Pre-processed tables are then picked up by the integrator script(s) and compiled into presentable tables.

Data sources

The pipeline integrates information about human and viral targets from the following sources:

Name	Source	Description
Human gene information	Ensembl	Information about human genes
COVID-19 UniProtKB	UniProt	UniProt site with information about SARS and SARS-CoV-2 proteins
SARS-CoV-2 Complexes	IntAct	Information about SARS-CoV-2 protein complexes
Human - virus interactome	IntAct	Human - SARS-CoV-2 interactome from Gordon et al. 2020 plus intercations of human proteins with proteins of other viruses based on IntAct data
Human interactome	IntAct	Human protein-protein interactions from IntAct database
Baseline expression per anatomical systems	Open Targets	Baseline gene expression per anatomical systems provided by Expression Atlas group used in the Open Targets Platform
Baseline expression distribution and specificity	Human protein Atlas	Information about subcellular location of proteins, tissue distribution and tissue specificity as provided by HPA
Protein expression during SARS-CoV-2 infection	Bojkova et al. 2020	Information about proteins whose abundance is regulated during viral infection from Bojkova et al. 2020 paper
Target Tractability	Open Targets	Target tractability assessment for small molecules, antibodies and other modalities provided by ChEMBL and used in the Open Targets Platform
Target Safety	Open Targets	Manually curated target safety data used in the Open Targets Platform
Target Drugs	Open Targets	Information about drugs extracted from the ChEMBL evidence file used in the Open Targets platform
Drugs in COVID-19 clinical trials	ChEMBL	Drugs in clinical trials against COVID-19
Active compounds in COVID-19 in vitro assays	ChEMBL	Compounds shown to be active in COVID-19 in vitro assays provided by ChEMBL
Mendelian randomization	Open Targets

Other files not listed in the table are also used for supporting purposes such as gene id mappings.

Usage

Prerequisites

The following programs have to be installed and available in order to run the pipeline:

• Python 3.7
• Pipenv: Recommended v2018.11.26 or newer. If using a package manager check the version installed, since version v11.9.0 available in Ubuntu does not work.
• jq
• R 4.0.0
• pandoc 1.12.3 or higher

The pipeline has been run successfully with those dependencies on macOS (Catalina) and Ubuntu (20.04 LTS).

Run project from scratch

git clone https://github.com/opentargets/ot_covid19
cd ot_covid19
make all

make all downloads all data, builds Python environment, run parsers and the integrator script(s).

Other actions:

• make setup-environment - building Python environment
• make downloads - download files only
• make parsers - run parsers only
• make integrator - run integrator(s) only
• make clean-all - removing temporary files.

Directory structure

Data folders:

• /data - containing version controlled data files that cannot be directly accessed from the web
• /temp - created by make will be populated when run locally. Data under this folder is not versioned
• /temp/raw_files - raw data files fetched from the web
• /temp/parsed_tables - parsed tables
• /temp/preformated_tables - pre-processed data ready to be integrated

Script folders:

• /src/parsers - contains parser scripts
• /src/query - contains SQL queries
• /src/integrators - contains integrator scripts

Data integration

For the integration part the concept is that we want to keep out logic altogether from the integrator scripts. All data processing steps should happen earlier in the parsing and pre-processing steps. Integrator scripts take tables with columns that are ready to be added to the final tables. So there are a few requirements:

1. Tables are read from /temp/preformated_tables directory only
2. The tables must be in an uncompressed tsv format
3. Tables must have a column called id containing unique identifiers (eg. Ensembl gene id or Uniprot primary accession)
4. Tables to be integrated must be added to the integration config files describing how the integration should happen

Integration configuration

This recipe shows how to integrate a dataset were we are expecting new targets (eg viral proteins) that are not included in the complete human gene set:

"uniprot_covid19_parsed.tsv": {
    "columns": [], 
    "flag": true, 
    "flag_label": "COVID-19 UniprotKB", 
    "how": "outer", 
    "columns_to_map": {
        "taxon_id": "taxon_id",
        "uniprot_ids": "uniprot_accessions"
    }
}

Where:

• columns contains the list of columns to be added. It can be empty if only flag is used.
• flag - boolean, indicating if the table is used as a flag eg. marking genes that are in the COVID-19 uniprot dataset
• flag_label - title of the flag column
• how - describing how the join should happen. By default it is left. Use outer if you expect new targets in the integrated dataset
• columns_to_map - column mapping to populate existing fields for the new targets.

This recipe shows how to integrate a dataset where we are not expecting new targets, and two new columns are added to the final table:

"ot_drugs_processed.tsv": {
    "columns": ["drug_names", "max_phase"], 
    "flag": false
}