Rhea, the reaction knowledgebase in 2022

1Rhea (https://www.rhea-db.org) is an expert-curated knowledgebase of biochemical reactions that uses the chemical ontology ChEBI (Chemical Entities of Biological Interest) (https://www.ebi.ac.uk/chebi) to represent reaction participants. Rhea covers enzymatic reactions and transport reactions, including but not limited to those described by the Enzyme Classification of the IUBMB, as well as reactions that occur spontaneously in biological systems. Rhea is now the reference vocabulary for enzyme annotation in the UniProt Knowledgebase UniProtKB (https://www.uniprot.org) and provides reaction data for a host of other resources including the enzyme knowledgebases IntEnz and the Enzyme Portal, the metabolomics data repository MetaboLights, the lipidomics knowledgebase SwissLipids, and the open chemistry database PubChem. Rhea reaction data is widely used for the annotation of genomes and genome-scale metabolic models, for integrated analysis of metabolomics data, and for computational pathway design.

2In this paper, we highlight a number of key developments in Rhea since our last publication in the database issue of Nucleic Acids Research. These include improved reaction coverage in Rhea and UniProtKB, the development of a new Rhea website, which provides powerful chemical structure and ontology search facilities to mine Rhea, and the selection of Rhea as an ELIXIR Core Data Resource.

3Rhea release 119 of June 2021 describes 13 673 unique reactions involving 11 886 unique reaction participants and evidenced by 15 500 unique literature references (PubMed identifiers), an increase of 2500 curated reactions, 1970 curated reaction participants, and 2889 curated literature references since our last publication on Rhea (which described release 96 of July 2018). Rhea reaction data is curated from the literature by expert biochemists, supported by natural language processing (NLP) tools such as LitSuggest, a web-based system for literature recommendation and curation. We have created a LitSuggest model that identifies literature relevant to enzymatic reactions, which we use to provide curators with a weekly digest of newly published literature and to scan all existing literature in MEDLINE, the National Library of Medicine's (NLM) bibliographic database. Tools like LitSuggest will be vital aids to filling our gaps in enzyme knowledge from the scientific literature.

4In 2019, the UniProt Consortium adopted Rhea as the reference vocabulary for enzyme and transporter annotation in the UniProt Knowledgebase UniProtKB. Rhea and UniProt curators now work closely together, with the Rhea editorial team curating new reactions in response to requests from UniProt curators, who then link those reactions to enzyme and transporter sequences in reviewed protein sequence entries in UniProtKB/Swiss-Prot. The UniProt automatic annotation resource, UniRule, provides Rhea reaction annotations for unreviewed protein sequence records in UniProtKB/TrEMBL. Reaction coverage in UniProtKB has grown by around 40% since the first publication describing enzyme annotation in UniProtKB using Rhea, up from 6654 Rhea reactions in UniProt release 2019_09 of October 2019 to 9294 reactions in UniProt release 2021_03 of June 2021 (around 68% of 13 673 Rhea reactions). UniProtKB currently provides Rhea annotations for over 222 000 UniProtKB/Swiss-Prot records and 23.2 million UniProtKB/TrEMBL records. These annotations power a range of enhanced chemical structure-based and chemical ontology-based searches over enzymes in UniProtKB via the UniProt website, API and SPARQL endpoint. Interested readers can find more details in the corresponding UniProt publications.

5In addition to UniProtKB, Rhea also provides reaction data for the SwissLipids knowledgebase, which features a library of over 700 000 known and theoretically feasible lipids that is fully mapped to the ChEBI ontology and community standard mass-spectrometry based lipid classifications. Rhea curators work closely with SwissLipids curators to capture knowledge of lipid metabolic pathways in ChEBI and Rhea (which SwissLipids curators then use to enumerate libraries of possible lipid structures in SwissLipids), and Rhea currently includes over 4600 reactions involving lipids. SwissLipids and Rhea curators are also working together to 'digitize' the SphinGOMAP resource of sphingolipid pathways (www.sphingomap.org) in ChEBI and Rhea. The SphinGOMAP, originally compiled by Professor Alfred Merrill Jr and collaborators, is an incredible resource of knowledge of sphingolipid pathways in graphical form; mapping this knowledge to ChEBI and Rhea makes it accessible for computation and facilitates the annotation of the corresponding enzymes in UniProtKB.

6The new Rhea website, released in October 2020, provides improved views of reaction data and powerful interactive search tools and programmatic access. We summarize the main features in the following sections, while users can also find further assistance at https://www.rhea-db.org/help.

7The Rhea reaction pages consist of several sections (Figure 1). The Reaction information section (Figure 1A) provides a graphical representation of the reaction (including chemical structures of reaction participants) served by a web component that developers can embed in their own webpages using our npm package (https://www.npmjs.com/package/@swissprot/rhea-reaction-visualizer). This section also includes information on enzymes in the form of links to the corresponding enzyme classes (EC number(s)), enzyme sequences (UniProtKB protein sequence records), and the corresponding Gene Ontology (GO) term (where a mapping of Rhea to GO is available from the GO Consortium). The Reaction participants section (Figure 1B) provides information on each reaction participant, including name, identifier, charge, formula, InChIKey (a simple hash representation of chemical structures that encodes information on connectivity, stereochemistry and charge in three distinct blocks) (https://www.inchi-trust.org), SMILES (Simplified Molecular-Input Line-Entry System) (http://opensmiles.org), a linear notation for chemical structures, and 2D structure coordinates (MDL Molfile). In addition, this section allows users to launch searches in Rhea for reactions involving each participant via a multi-faceted tooltip linked to the name. The Cross-references section (Figure 1C) provides links to relevant information from a range of resources including reactions from KEGG, Reactome, MetaCyc, EcoCyc, and M-CSA, and enzyme information from UniProtKB, the Enzyme Classification, and the GO, and indicates to which member of the quartet of Rhea reaction identifiers each cross-reference applies (Figure 1C). The remaining sections are Related reactions, which provides a list of reactions that are either general or specific forms of the current reaction (parent/child reactions), Publications, which lists the peer-reviewed literature from which the reaction was curated, and Comments, which provides additional information about the reaction. The Rhea reaction page also features action buttons (at the top of each page) that allow users to copy a textual representation of the reaction equation on a clipboard or download the reaction in RXN or RD format.

8Users can search Rhea by providing search strings that may include chemical names, chemical identifiers (ChEBI, Beilstein & CAS numbers) and reaction identifiers (KEGG, Reactome, MetaCyc, EcoCyc and M-CSA), enzyme classes (EC numbers), UniProtKB accession numbers, GO term identifiers, and PubMed identifiers. Clicking on Advanced search opens the query builder (Figure 2A), which allows users to select specific search fields (which is useful for disambiguation) and to combine searches in fields using Boolean operators AND, OR, NOT (Figure 2B). The query builder also provides an autocomplete feature for some fields (ChEBI names and GO Molecular Function terms) that helps users to choose among a number of related possibilities as they type. We provide an exhaustive list of the query fields used in Rhea in Supplementary Table 1.

9Certain search fields in Rhea contain terms from ontologies such as ChEBI and the GO. Rhea indexes the ‘is a’ relationships from both ChEBI and the GO, such that a user who searches Rhea using an identifier from either ontology will be presented by default with all reactions involving (or mapped to) members of that class, and of any child class. Hence, a search for ChEBI:18059, the class of lipids, will return all reactions involving any type of lipid. Users who wish to override this default behavior, and search using ChEBI and GO identifiers without considering the ‘is a’ relationships, can do so by selecting the Exact option in the Advanced search interface (Figure 2C). Users can also search for all reactions that map to a specific level of the enzyme classification using partial EC numbers, such as ‘ec:2’, ‘ec:2.1’ or ‘ec:2.1.1’.

10The following query illustrates a number of the aforementioned features, combining multiple fields using Boolean operators, searching ontologies and hierarchical vocabularies, and using resource names, identifiers, and wildcards. This query retrieves all reactions annotated in UniProtKB that involve lipids, that map to EC class 1 (Oxidoreductases) and that have a mapping to the GO and Reactome (Figure 2C and D):

11UniProt:* AND ChEBI:18059 AND EC:1 AND GO:* AND Reactome:*

12We close this section with a brief description of how Rhea handles protonation states in ChEBI. The ChEBI ontology treats each protonation state of a given chemical entity as a distinct entity, with its own unique identifier. Rhea selects only one protonation state for each chemical entity by default, that which represents the major protonation state at pH 7.3. The Rhea search engine maps ChEBI identifiers for other protonation states to the major protonation state used in Rhea, using a precomputed mapping of chemical structures available at https://www.rhea-db.org/help/download. This applies to Exact searches too, the logic being that if other protonation states were not mapped there would be no results.

13Users can search for Rhea reactions using complete or partial InChIKeys through the Advanced search or the simple search interfaces (using the prefix ‘inchikey:’). Users can also map lists of (complete or partial) InChIKeys to Rhea reactions using the bulk retrieve/ID mapping functionality described below.

14The Rhea website supports searches for reactions that involve a compound that is either similar to, or a derivative of, a given compound of interest, via the structure search interface at https://www.rhea-db.org/structuresearch. Rhea uses the IDSM Sachem chemical cartridge for both fingerprint-guided similarity and substructure searches. Users can provide query structures in the form of valid SMILES or by importing or drawing structures using the Ketcher molecular editor provided (https://lifescience.opensource.epam.com/ketcher).

15The Rhea website presents search results in tabular form, where each row corresponds to a single reaction and each column corresponds to one aspect of a particular reaction (Figure 3). Action buttons allow users to add or remove columns from the result table, to download all or selected results in a range of formats, to map a selection of the reactions to UniProtKB, and to create a link for the result that can be bookmarked, shared and reused (see section ‘Programmatic access’ below). Users can apply additional constraints to the result set by clicking on filters in the left side navigation panel, limiting the results to specific types of reaction participants (such as proteins), specific types of reactions (such as transport reactions), or specific enzyme classes. A Refine search feature lists relevant fields that users can select when using query strings that match multiple fields. The chemical similarity and chemical substructure searches do not return a list of reactions but rather a list of ChEBI compounds that match the query structure and that are participants in Rhea reactions.

16The Rhea identifier mapping service (https://www.rhea-db.org/mapping) accepts identifiers for reactions from KEGG, Reactome, MetaCyc, EcoCyc and M-CSA, identifiers for chemical entities (currently ChEBI identifiers and the InChIKey), and identifiers for enzyme classes and functions (currently EC numbers and identifiers from the Molecular Function branch of the GO). Users can copy-paste a list of identifiers, or upload a file containing a list of identifiers, and specify the input identifier type, in order to obtain a mapping of their identifiers to Rhea reactions. As with the Advanced search feature, hierarchical mapping is performed by default for GO and ChEBI, in which a user specified term is mapped to all Rhea reactions mapped to child terms in the relevant ontology (Figure 4A), but the drop-down menu also provides an Exact mapping option (Figure 4B). The Rhea identifier mapping service provides results in the standard result table, with an additional column for each row that lists the user-provided identifier(s) that were mapped to that Rhea reaction (one Rhea reaction per row). The identifier mapping tool also accepts Rhea reaction identifiers, which allows users to retrieve a list of Rhea reactions in tabular format.

17Rhea provides programmatic access to all data, queries and tools available through the Rhea website via RESTful URLs that users can bookmark, link, and use in their own programs. The easiest way to create a URL is using our interactive Advanced query builder, and then use the Share icon to obtain the URL to which the requisite format parameter can be added. Individual reaction data is available in RXN and RD formats, with search results in customizable tab-separated formats. Supplementary Table 2 provides the possible query parameters for a REST request and Supplementary Table 3 provides the list of possible columns to customize the result set. The examples shown below - for Unix and Python 3 - use the REST API to retrieve all Rhea reactions mapped to UniProtKB enzyme sequences and to export them as a tab-delimited file containing Rhea reaction identifiers, reaction equations, and UniProt accession numbers.

18The Rhea SPARQL endpoint (https://sparql.rhea-db.org/sparql) supports complex and federated queries over Rhea RDF and RDF of other resources providing SPARQL endpoints through federated queries, including chemical similarity and substructure searches using the IDSM/Sachem SPARQL endpoint. We invite interested readers to consult the SPARQL endpoint and sample queries and documentation provided there.

19ELIXIR (https://elixir-europe.org) works to link national centres and core bioinformatics resources in Europe into a single coordinated infrastructure for life science data. ELIXIR has created a formal process to identify the most critical life science data resources in Europe, which are termed ELIXIR Core Data Resources (CDRs). ELIXIR selected Rhea as an ELIXIR Core Data Resource in 2021. There are currently 22 ELIXIR Core Data Resources (CDRs), covering genes and genomes, proteins, small molecules, molecular structures, interactions, and literature. Rhea is the first ELIXIR CDR that focuses on reactions and bridges UniProt, the ELIXIR CDR for proteins, and ChEBI, the ELIXIR CDR for small molecules.

20Rhea is a reference resource of computationally tractable enzyme and transport reaction data and the standard for enzyme and transporter annotation in UniProtKB. The Rhea website, API and SPARQL endpoint provide a powerful toolbox to mine the Rhea reaction dataset for a broad range of applications, while the UniProt website, API and SPARQL endpoint allow users to exploit the rapidly growing set of enzyme and transporter annotations created using Rhea. Perhaps one of the most exciting applications of Rhea (and UniProt) data to emerge recently is the use of machine intelligence to study and design enzymes and biosynthetic and bioremediation pathways—combining state of the art deep learning models of language to describe both protein sequences and small molecule chemistry. We will continue to develop the Rhea dataset, website and modes of programmatic access to better support these and other applications, and will continue to work with other key knowledge resources such as UniProt, the GO and Reactome to improve the consistency and interoperability of enzyme, transporter, and reaction data in all of our resources.

21The Rhea website is available at https://www.rhea-db.org/ and the Rhea SPARQL endpoint at https://sparql.rhea-db.org. All data in Rhea is freely available under a Creative Commons Attribution License (CC BY 4.0) and users can download it from our FTP site https://ftp.expasy.org/databases/rhea/ in the following formats: RDF, BioPAX, RXN/RD, and TSV. More information regarding different downloadable files provided by Rhea is available at https://www.rhea-db.org/download. We now synchronize Rhea and UniProtKB releases (as of February 2020), which are published approximately every eight weeks. We provide a snapshot of the ChEBI data corresponding to the Rhea and UniProtKB release on the Rhea ftp site, as well as an export of the 2D structures in MOL and SDF formats. Video tutorials on how to use Rhea website are freely available on the SIB (Swiss Institute of Bioinformatics) YouTube channel https://www.youtube.com/channel/UCPo4ED_WAKjwQ878cca6_oQ - we list available videos in Supplementary Table 4. Users that would like to have reactions added to Rhea are very welcome to use the Feedback form (https://www.rhea-db.org/feedback) - input from our users is highly appreciated.