DGIdb 5.0: rebuilding the drug-gene interaction database for precision medicine and drug discovery platforms

Abstract

The Drug-Gene Interaction Database (DGIdb, https://dgidb.org) is a publicly accessible resource that aggregates genes or gene products, drugs and drug-gene interaction records to drive hypothesis generation and discovery for clinicians and researchers. DGIdb 5.0 is the latest release and includes substantial architectural and functional updates to support integration into clinical and drug discovery pipelines. The DGIdb service architecture has been split into separate client and server applications, enabling consistent data access for users of both the application programming interface (API) and web interface. The new interface was developed in ReactJS, and includes dynamic visualizations and consistency in the display of user interface elements. A GraphQL API has been added to support customizable queries for all drugs, genes, annotations and associated data. Updated documentation provides users with example queries and detailed usage instructions for these new features. In addition, six sources have been added and many existing sources have been updated. Newly added sources include ChemIDplus, HemOnc, NCIt (National Cancer Institute Thesaurus), Drugs@FDA, HGNC (HUGO Gene Nomenclature Committee) and RxNorm. These new sources have been incorporated into DGIdb to provide additional records and enhance annotations of regulatory approval status for therapeutics. Methods for grouping drugs and genes have been expanded upon and developed as independent modular normalizers during import. The updates to these sources and grouping methods have resulted in an improvement in FAIR (findability, accessibility, interoperability and reusability) data representation in DGIdb.

Graphical Abstract

Figure 1.

Architecture and data access for DGIdb v5.0. (A) DGIdb v5.0’s data are housed in a PostgreSQL database supported by Ruby on Rails version 6.1.3. Data from DGIdb is accessible via an HTML GUI built with React.js (http://dgidb.org) or through a GraphQL API (https://dgidb.org/api/graphql). (B) DGIdb’s web interface has been redesigned using the React.js framework. All previous functions and search types (interactions by drug, interactions by gene, gene categories) are accessible through the new interface (http://dgidb.org). (C) DGIdb supports GraphQL queries through the new API page. Queries can be sent here via API calls (https://dgidb.org/api/graphql) or sent manually through our GraphQL playground widget. Further documentation and example queries can be found at http://dgidb.org/api.

Figure 2.

New and updated sources in DGIdb v5.0. Several new sources have been added to DGIdb and many existing sources have been updated. (A) Five new drug claim sources have been added and many existing sources updated. New drug claim sources include ChemIDplus, Drugs@FDA, HemOnc, NCIt and RxNorm. (B) HGNC has been added as a new gene claim source and many existing sources have been updated. Abbreviations: CF, Clearity Foundation; CGI, Cancer Genome Interpreter; CIViC, Clinical Interpretation of Variants in Cancer; DoCM, Database of Curated Mutations; DTC, Drug Target Commons; FDA, Food and Drug Administration; FO, Foundation One; HGNC, HUGO Gene Nomenclature Committee; IDG, Illuminating the Druggable Genome; JAX-CKB, The Jackson Laboratory Clinical Knowledgebase; MCG, My Cancer Genome; MSK, Memorial Sloan Kettering; NCI, National Cancer Institute; NCIt, National Cancer Institute Thesaurus; TALC, Targeted Agents in Lung Cancer; TDG, The Druggable Genome Clinical Trial; TEND, Trends in the Exploration of Novel Drug Targets; TTD, Therapeutic Target Database.

Figure 3.

Record counts for drug, gene and interaction claims within DGIdb (v4.0 compared against v5.0). The number of individual records for each record type (drug, gene, interaction) is shown. The number of drug records has increased, while records for genes and interactions have been trimmed to just normalized records, resulting in an increase in data quality for the end user.

Figure 4.

Normalization updates for drug and gene records across DGIdb (v4.0 compared to v5). (A) Drug normalization using thera-Py shows improvement in rates of normalization from 68.82% of records in v4.0 to 88.58% of records in v5.0. (B) Gene normalization using the VICC Gene Normalizer shows improvements in rates of record normalization of 76.09% in v4.0 to 94.91% in v5.0.

Figure 5.

Addition of GraphQL enables ad hoc queries in DGIdb. GraphQL playground interface and example queries enable users to write their own custom queries for aggregate interaction and categories data. (1) To access the GraphQL API playground page, click the API footer button on the main page. (2) GraphQL query panel. All queries to be run can be written and edited within this panel. (3) GraphQL results panel. Results from queries written in panel 2 will populate here. (4) Documentation and query history panel. Documentation for all supported query types and data points can be accessed from here. Similarly, past queries are saved and can be rerun as needed. (5) Query control functionality. Once written, queries can be executed, prettified and merged using the buttons available. (6) Demo queries panel. Example queries are provided for commonly used search types. Queries can be copied into panel 2 and run as written, or edited further for additional data points. Additional examples of ad hoc queries have been provided in Supplementary Figures S2 and S3.

Figure 6.

New UI expands search and filter control for interaction searches. Implementation of ReactJS into DGIdb has enabled a complete redesign of existing UI elements. (1) Search type selection. Desired search type can be selected from the main page dropdown: interactions by gene, interactions by drug and gene categories. (2) Search field. Once all drugs or genes of interest are entered, the search button can be used to submit the search request. (3) Match type tabs. All unique matches for requested terms will populate the ‘Unique Matches’ tab. Terms that are ambiguous (have multiple potential matches) or do not return any results will populate the ‘Ambiguous or Unmatched’ tab. (4) Gene summary panel. A summary of unique results for each entered search term will populate this panel. ‘At-a-glance’ infographics are provided for different metrics of the result set. Individual search term labels can be selected to filter both the infographics and results table. (5) Interaction results table. Unique results for searched terms will populate this table. Results contain all: searched term, gene name, drug name, regulatory approval status, known indications and interaction score. This table can be filtered using the search terms in panel 5 or using additional filter controls shown in panel 6. (6) Additional filter controls. Dropdowns on each column in the results table can be filtered with additional options. These controls allow filtering for specific strings of interest using common relational operators, such as ‘contains’, ‘equals’, ‘starts with’, ‘ends with’, ‘is empty’, ‘is not empty’ and ‘is any of’. (7) Download results. Results from search can be downloaded and analyzed externally from DGIdb. Results are downloaded in a TSV format.