MIBiG 4.0: advancing biosynthetic gene cluster curation through global collaboration

1Many organisms are prolific producers of small molecules known as specialized or secondary metabolites (SMs). These molecules often show a diversity of potent biological activities, which have been leveraged for the development of numerous drugs. SMs are generally hypothesized to increase the fitness of the producing organism or its host. In microbes, the biosynthetic genes required for the production of an SM are co-regulated and frequently physically clustered in the genome, in a so-called biosynthetic gene cluster (BGC), and often transferred horizontally. BGCs, which by definition consist of two or more genes, encode the proteins/enzymes used in biosynthesis, resistance and regulation of SMs and are the object of ‘genome mining’ strategies that leverage analysis of genome sequence data for the discovery of (novel) metabolites.

2Over the last decades, various methods using manually curated detection rules based on prior knowledge, and more recently, machine learning-based tools for genome mining have been developed. These tools rely on accurately curated and machine-readable experimental data for annotation, rule definition and training purposes. Unfortunately, machine-readable data are neither readily available from the scientific literature nor universally required by journals to be directly deposited in databases. While there are efforts to mine data from the literature using computational methods, these approaches currently often come with limitations when compared with human curators and may not be compatible with copyright laws. Therefore, manual data curation performed by researchers remains the gold standard for the generation of machine-readable data.

3The largest manually curated resource on SM BGCs is the Minimum Information about a Biosynthetic Gene Cluster (MIBiG) data repository. Initiated in 2015 and based on the MIBiG Data Standard, it now holds over 2500 hand-curated entries of experimentally validated BGCs and their products, alongside additional information such as biological activities and gene annotations. In rare cases, a single gene may be responsible for the biosynthesis of a natural product, such as a large non-ribosomal peptide synthases; these standalone genes are also entered into MIBiG due to their relevance to specialised metabolism. Conceptualized as an open data repository curated by and for the SM community, the MIBiG repository has seen three iterations of online community-driven data annotation and curation hackathons (also known as ‘annotathons’), with >250 participants from 33 countries. Despite its size, the MIBiG repository still only describes a part of the continuously growing known biosynthetic space, which motivates further efforts in curating and systemizing information on BGCs.

4Here, we present version 4.0 of the MIBiG data standard and repository. Besides a thorough update of the underlying MIBiG data standard, we have substantially grown the number of available entries by initiating a large-scale community curation effort. In the first half of 2024, 267 contributors created 557 new entries and modified 590 existing entries in the scope of eight community annotathons (six general open events and two final data curation sessions with a more dedicated team). In this version of MIBiG, we focused on maintaining and further improving data quality in terms of completeness and accuracy. We encouraged contributors to fully complete entries before submission, which has significantly decreased the number of so-called minimum entries (entries with only the minimally required information) in the database. We also introduced a new peer-review model where modifications to entries are examined and approved by one or more volunteer expert reviewers, who can request corrections from data submitters. Additionally, we have established an initial prototype for efficient and standardized data submission, and during the annotathons we utilized a web interface (MIBiG Submission Portal) that allows for parallel, distributed data input featuring automated input validation. The latter refers to the tests that are performed by the submission portal itself to ensure the correct data types and formats are filled in. Together, these efforts further consolidate MIBiG as the leading database on experimentally characterized BGCs and prepare for the transition to a dynamic, rolling-release curation model.

5The MIBiG Data Standard (from here onwards, Data Standard) is the ‘blueprint’ of all allowed data in the MIBiG repository. It defines mandatory and optional data fields, allows the use of controlled vocabularies and automated validation and enables the organization of complex data in a consistent, human- and machine-readable way. In this update, we extensively revised the Data Standard to accommodate advances in SM research and to extend the scope and ease of (re-)use of covered (meta)data.

6Previously, all literature references associated with a MIBiG entry were collected in a single block, making it difficult to locate the origin of specific experimental data. In this update, we reorganized the Data Standard such that each data category (e.g. biosynthetic information, compound details, etc.) has its own list of literature references. Furthermore, evidence qualifiers can be selected from a controlled vocabulary (e.g. ‘heterologous expression’) that concisely summarizes the experimental support for the claims. While newly added entries adhere to these changes, entries added in previous versions of MIBiG still follow the legacy format, and will be updated gradually over time. To summarize the data quality of an entry concisely, we also introduced a ‘Quality’ identifier, and it is possible to filter entries based on high, medium or questionable quality of data. Note that this label only reflects the presumed data quality of an MIBiG entry and does not address the quality of the underlying literature.

7Biosynthetic information is now organized in a ‘biosynthesis’ section, tracking biosynthetic types, modules, operons and newly introduced ‘biosynthetic path’, which allows contributors to describe cases where a single BGC can lead to multiple products or describe sub-clusters of genes that produce building blocks. The ‘multiple loci’ system has been re-introduced, allowing the specification of satellite genes or gene clusters that are involved in the biosynthesis but are not clustered with the ‘main’ BGC. Nevertheless, we still require that multiple biosynthetic genes are clustered in the same genomic region, to exclude non-clustered pathways. Furthermore, it is now possible to mark genes that are located within the boundaries of a BGC but do not partake in the biosynthesis, such as pseudo-genes or transposable elements. Additionally, we have separated biosynthetic classification from compound classification (e.g. we removed ‘alkaloid’ as a biosynthetic class) and introduced a custom biosynthesis-inspired chemical ontology for SMs (Supplementary Data 1, section 3.4) based on the work by Dewick. Furthermore, we have newly defined the non-ribosomal peptide synthetase Type VI (modular, non-condensation-domain peptide-bond-forming), extending the current classification.

8MIBiG also accepts additional BGC-related data. In this update, we have reworked fields registering the biological activity of BGC-associated SMs: activities are now considered properties of a specific assay, and a controlled vocabulary (Supplementary Data 1, section 3.3) is available for defining bioactivity in a reproducible way. Additionally, we have included an optional ‘Concentration’ field, allowing submission of both qualitative and quantitative bioactivity data. At the same time, additional metadata parameters increase the scope of the already extensive Data Standard, and as such MIBiG references external resources where possible. Newly introduced links include references to the Minimum Information about a Tailoring Enzyme (MITE) data repository for annotation of tailoring enzyme-encoding genes, and CyanoMetDB for compound information on cyanobacterial SMs.

9Inspired by the contributions made to MIBiG 3.0, we again sought participation from the scientific community. Following calls on social media, 398 researchers signed up to participate in a series of eight 3-h online annotation sessions, accommodating different time zones (Figure 1). This enormous interest posed organizational challenges in terms of coordination and communication, prompting us to develop a new model for community participation. Individual contributors were part of one or more Interest Groups that communicated using the MIBiG Slack (https://mibigannotathons.slack.com/) channel and were headed by Interest Group Coordinators: topic matter experts responsible for answering biosynthesis- and chemistry-related questions. Kanban-style boards (free version of Trello, https://trello.com/) were employed to coordinate work on entries. Data submission was performed using a MIBiG Submission Portal prototype, a bespoke web interface that uses validated fields for data processing (code available at https://github.com/nlouwen/submission-prototype). Several curators with relevant expertise volunteered to take Reviewer roles, focusing on assessing the quality of newly generated or modified entries using the newly introduced peer review system. Aimed towards further improving the quality and confidence of entries, Reviewers could leverage the Kanban-style boards (Figure 2) to request revisions of entries if errors were found. To facilitate data curation, we prepared extensive online documentation (Supplementary Data 1) and instructional videos, and trained Interest Group Coordinators and Reviewers for their roles in online meetings. Participants who made a significant contribution (defined as participating in at least two 3-h sessions or an equivalent time investment) were invited to be co-authors in the present publication.

10In this iteration of the MIBiG annotathons, we put a greater emphasis on self-organization and facilitating motivated contributors to act independently and confidently when curating data. During the call for participation, researchers not only signed up to participate, but also contributed to assembling a list of recent publications associated with the biosynthesis of SMs. This initial effort yielded 552 publications supporting new entries and 266 publications for improvements of existing entries, which were used as a starting point for the curation process. Over the course of the annotathons, 267 contributors made a total of 8304 edits (e.g. adding an entirely new entry, adding biological activity to an existing entry, etc.), resulting in 557 new and 590 modified existing entries. With the present update, MIBiG now contains a total of 3059 entries, a 22% increase in comparison to MIBiG 3.0. Of these, 1655 entries are now associated with 3604 biological activities, and 2634 entries have 5002 associated chemical structures. However, 672 entries still lack chemical structures; hence, future efforts will include attention to improving this aspect, especially with regard to structural information for ribosomally synthesized and post-translationally modified peptides. Additionally, 7677 references and 8582 evidence qualifiers were provided, 171 biosynthetic paths were described for 110 entries and cross-references to 173 MITE and 93 CyanoMetDB entries were established. A summary of the changes in comparison to MIBiG version 3.1 can be seen in Figure 3.

11Of the total 1147 contributed entries (557 new, 590 modified), 464 (40%) have been reviewed at the time of manuscript preparation. While all entries are available, those that are reviewed are highlighted in the MIBiG repository website to reflect the additional confidence. For applications using the MIBiG data where a high confidence level is required (e.g. machine learning applications), we recommend the use of reviewed entries only (the website facilitates filtering/sorting on this). We expect the ‘reviewed’ part of the MIBiG repository to grow continuously once we have transitioned to the MIBiG rolling release model, and over time, we aim to formally review all entries in the MIBiG repository.

12The aforementioned efforts demonstrate the value of leveraging large community initiatives such as the MIBiG annotathons. We estimate that contributors volunteered ∼4000 h in curating and reviewing entries, an effort in time and expertise that could not be raised by any single research group. Besides expanding the MIBiG repository, the annotathons were appreciated for their community-building aspect, fostering communication and exchange of ideas in the SM research community. In addition, the interaction with other resources prompted improvements to these databases as well, e.g. when curators could not find matching entries for a structure in the NP Atlas, thus encouraging wider cooperation beyond MIBiG itself. The broad interest of the community motivated the planning of a ‘rolling release’ model of MIBiG. In addition to the biennial efforts that will lead to ‘major’ releases of MIBiG (e.g. the current v4.0, or the next major release v5.0), curators will be able to contribute new or modify existing entries on an ad hoc basis, leading to quarterly ‘minor’ releases (i.e. 4.1 and 4.2). Contributors will be able to correct bugs and add references at any time, instead of waiting for the ‘major’ release cycle to perform all edits at once. This new system is currently under development, and we invite the scientific community to participate in the discussion on how to structure contributions and governance (i.e. by communicating with the corresponding authors of this publication or using the MIBiG Slack Workspace https://mibigannotathons.slack.com). Furthermore, to facilitate future MIBiG updates and curation we encourage authors to release BGC sequence data during the publication submission and peer review process, or immediately thereafter, and to provide the respective accession details in the manuscript text.

13In summary, we have conducted a large-scale community effort to make experimental data on SM BGCs freely accessible and machine-readable. As a resource created for and by the scientific community, the MIBiG repository is freely accessed on an entry-by-entry basis or can be downloaded and parsed in bulk. MIBiG 4.0 also serves as the stepping stone for creating the infrastructure to establish a Wikipedia-like model of continuous community curation. Such a decentralized organization will guarantee continuous development of MIBiG and help in including the next generations of scientists in the annotation and development process.