Ten Simple Rules for Developing Public Biological Databases

Biological databases are online libraries that contain structured information about living organisms. These databases are indispensable research tools, as they provide convenient, computable access to prior knowledge that is vital for planning future experiments and for discovering new knowledge through data mining—they help us “stand on the shoulders of giants.” Because of their importance to research, the number of public biological databases is increasing. For instance, the number of biological databases published per year in the journal Nucleic Acid Research (NAR) increased dramatically from only two databases in 1980 to 182 in 2016, with the expectation that this single journal will have published over 2,500 database articles by the end of 2017 [1]. Some of these databases are key, sophisticated, user-friendly, long-term, stable resources, built and maintained by professional teams. However, others have been criticized for being difficult to use or having unclear data quality levels [2,3], and many become obsolete over time [4]. So, if you are considering developing a new database, and especially if you are a student or postdoc, please, for the love of science, follow these ten simple rules for creating and maintaining biological databases (and also a similar set of great rules for scientific web resources [5]). Rule 1: Don’t reinvent the wheel Creating a high-quality database is a responsibility that involves strong commitment to accurate data collection and regular content and feature updates, not to mention a substantial time investment. The strongest reason to create a new database is scientific demand for a type of data not easily available in a computable form anywhere else. It is most useful to have all data of a single type in one easy-to-search location, so, ideally, everyone interested in collecting data about a specific topic should collaborate to create one resource, or at least should coordinate efforts to reduce duplication of work (Fig 1). Either way, the data content and software features you create will have the greatest impact if they are original and useful; thus, a comprehensive literature review is a necessary starting point that comes before any actual work (as in any scientific endeavor). A good place to start searching for relevant prior work is the “NAR Online Molecular Biology Database Collection,” which, as of January 2016, contains 1,666 biological databases organized into categories [1], collected from the annual NAR database issue. NAR also publishes an annual web server issue dedicated to web-based software resources [6], and the journal Database focuses on biological databases and curation. Several online directories maintain biological database link collections, such as Pathguide (547 databases) [7], The Tools and Data Service Registry (557 databases) [8], The Bioinformatics Links Directory (623 databases) [9], OMICtools (1,513 databases) [10], and MetaBase (1,802 databases) [11]. Open in a separate window Fig 1 How biological databases proliferate (adapted from https://xkcd.com/927/ and drawn using Comix I/O [http://cmx.io/]). Rule 2: The three most important things in database development are data quality, data quality, and data quality Some databases collect unique content directly from experiments via authors (e.g., GenBank [12]) or expertly curated from the literature (here, we call these “primary”), and some collect non-unique content from other databases (we call these “secondary” or meta-databases). Primary data could be collected from a community of data generators, such as with GenBank [12], or curated by experts, such as with UniProtKB/Swiss-Prot [13]. Secondary databases, such as InterPro [14], host data collected from other public resources that are often formatted and processed in a uniform way, which can create a more comprehensive and useful data source. Depending on the database type, there are different standards for quality control. Primary databases are responsible for thoroughly checking the quality of all input data, for instance using manual cross-checking and rule-based automated data entry validation [15,16]. It is a good idea to develop a manual describing standard operating procedures for curation to help maintain and communicate quality standards. Secondary databases must ensure to not introduce new errors via their data integration process. Ideally, the integration process will improve data quality via cleaning and normalizing (e.g., standard database identifier use). In either case, following best practices and standards can help improve data quality. A data model standard may require that particular information (e.g., database or gene identifier, an ontology) is used, otherwise intended uses will not be possible. Often, an automatic data validator is available for a standard format to ensure that rules are followed, leading to more consistency and fewer errors. For meta-databases, transforming all incoming data to a standard format eases data integration and can reduce software development time and implementation errors, as well-tested software libraries are available to help read, write, and access the data. For example, the Proteomics Standards Initiative Molecular Interaction (PSI-MI) format captures molecular interactions, such as protein–protein interactions, and maintains a validator, a software library, and other tools to ease working with this data type [17]. Data provenance, information about where the data has come from and when it was updated, is also important to capture and track in order to help users critically evaluate database content quality.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.