香京julia种子在线播放

In the US National Academies of Science, Engineering and Medicine workshop on “Changing the Culture on Data Management and Sharing (Martone and Nakamura, 2022), one of the main takeaways was that the focus of data sharing efforts should not be targeted toward the individual investigator, but the laboratory. As one participant noted: “If you can share data with people in your lab, you are much more likely to have something worthwhile to share outside the lab.” FAIR data management is therefore an intentional lab-wide strategy that ensures that data can be shared with lab mates, the PIs, and other colleagues, your future self and eventually with the broader scientific community. Across all stages of the data lifecycle, the management strategy puts in place processes so that data can be found, accessed, combined when necessary, and reused. By paying attention to FAIR in the laboratory throughout the life cycle, benefits start to accrue to the data creator, the laboratory, PI, and collaborators well before data flows out to the wider scientific community (Bush et al., 2022; Dempsey et al., 2022).

Examples of lab management practices built on the FAIR principles are given in Table 2.

We are starting to see neuroscience researchers sharing their experiences with developing and utilizing lab-centric data management systems. They range from tightly integrated digital infrastructures (Bush et al., 2022; Dempsey et al., 2022) to a set of practices that can be implemented using “off the shelf” components for an average neuroscience wet lab (Fouad et al., 2023).

One of the most important steps for a researcher in ensuring that their data is FAIR for the long term is to submit their data to a trustworthy repository that supports FAIR. The new NIH data sharing policy requires researchers to indicate where they will be sharing their data as part of the data management and sharing plan. As recommended in Table 2, knowing in what repository the data will be published allows the researcher to understand what standards are required so they can be built into the laboratory management workflow. With its growing ecosystem of specialized databases, researchers have a choice about where to publish their data.

Understanding how the neuroscience repository landscape is organized may help in finding the right repository. Repositories are generally specialized by data type (Figure 2). However, repositories also exist that are specialized for a domain, e.g., the SPARC database accepts all data associated with the peripheral nervous system, or serve researchers within a particular region, e.g., CONP, or institution, e.g., BrainCode and the Donders Repository. Often, multiple repositories may be appropriate, in which case there are additional features that may make a given repository more or less attractive. These include tool support, curation services, support for data citation, choice of license, size of data allowed, help with data management plans (see Table 2) and possible costs (Murphy et al., 2021). A functioning neuroscience ecosystem also requires open neuroscience repositories that have few restrictions on data types, regions, or subdisciplines to ensure that all data has a home. The EU EBRAIN infrastructure is an example of such a repository, as it takes multiple types of data regardless of discipline or geographical location, although there may be issues with transferring certain types of data across international borders (Eke et al., 2022).

Supplementing the specialist repository landscape are the generalist repositories, data repositories that span scientific disciplines and data types (Assante et al., 2016). These repositories are often useful for publishing smaller supplemental datasets that are required for a publication (Stall et al., 2023). Specialist repositories generally provide more standards, tools and services for harmonizing and using data, and make it easier for researchers to find data of a particular type. To aid researchers in choosing an appropriate neuroscience data repository, the INCF has a searchable infrastructure catalog, where each repository is described according to the checklist developed by Sandström et al. (2022). Other repository finder tools include NITRC for neuroimaging related repositories, re3data, the catalog of open data repositories maintained by the National Library of Medicine, and the NIF listing of BRAIN Initiative Repositories.

While the investigator takes the central role in acquiring data in a manner that supports FAIR, the community repository is arguably the central player in implementing the basic requirements for achieving FAIR for the long term (Figure 1). We are using the term “community repository” here to designate infrastructures that are designed to accept primary data contributed by outside researchers, rather than a single data set produced by a given project (e.g., the Allen Brain Atlas) or a knowledge base that aggregates information about a particular entity (e.g., CoCoMac).¹ As shown in Figure 1, the repositories have critical responsibilities for ensuring that submitted data are made available according to the FAIR principles (Lin et al., 2020). These practices include issuing and maintaining persistent identifiers, tying those identifiers to rich metadata, providing access and any necessary access controls, enforcing or supporting annotation with FAIR vocabularies, enforcing or supporting community standards, supporting data versioning, providing links to other critical products like experimental protocols and code, and provisioning a clear data license for each data set. Repositories also have the critical role of ensuring that data is available for the long term.

From the earliest days of neuroinformatics, it was envisioned that neuroscience would likely best be served by a decentralized system of federated databases (Koslow, 2000). Due to the variety and complexity of neuroscience data, a single large repository like Genbank or the Protein Data Bank was likely not going to be feasible. The early investments in neuroinformatics by the US Human Brain Project and the success of the International Neuroinformatics Coordinating Facility in growing the field of neuroinformatics globally, led to the first generation of neuroscience databases. These databases were largely organized around data type, e.g., structural neuroimaging (XNAT), functional neuroimaging (fMRI Data Center; Open fMRI), neurophysiology (CARMEN; Neurodatabase.org,” GNode), EEG (open EEG, iEEG), neuronal morphology (NeuroMorpho), microscopic images (Cell Centered Database), neuromodeling (ModelDB). Some examples are shown in Table 1.

When the first generation of neuroscience databases were started, there were few standard practices for designing web-accessible databases. As documented by NIF, each database had a different mode of access, different data structure, and the use of standards was very limited. It was a time of tremendous technological fluidity, with standard features we take for granted today (e.g., RESTful web APIs) still being invented. The cloud did not exist, and attempts to build resources on the early version of a cloud-like system (“the grid”) met with considerable challenges (Grethe et al., 2005). With today’s emphasis on data sharing, increased attention is starting to be paid to these critical infrastructures and how they are constructed, operated, and evaluated (Nelson, 2022). Various recommendations on desired characteristics for data repositories have been issued by different groups (Sansone et al., 2020; Shearer, n.d.), including NIH (Selecting a Data Repository) and additional sets of principles, e.g., the TRUST principles (Lin et al., 2020) and principles for open infrastructures (Bilder et al., 2015) have been formulated to help further guide how these critical infrastructures should operate. The Elixir project, a large scale bioinformatics consortium in the EU, has developed a maturity model for evaluating the success of repositories which is designed to be used by funders to determine the criticality of various infrastructures (Bahim et al., 2020). The INCF Infrastructure working group recently issued a set of guidelines from a neuroscience perspective, that provide a mix of technical and “customer service” recommendations for operating repositories (Sandström et al., 2022). Although these various lists of desiderata do not overlap completely (Murphy et al., 2021), over time we will likely converge on a core set of functions and expectations for these critical infrastructures, balancing the often dual requirement for these infrastructures to serve as both publishing platforms and dynamic scientific gateways (Sandström et al., 2022).

INCF has served as an important conduit by which the FAIR principles have permeated the construction of neuroscience data repositories and gateways. Investigators who have been active in INCF through governance, committees and working groups are involved with several of the next generation neuroscience infrastructures including EBrains, CONP, SPARC, DANDI, Open Neuro, and BRAIN/Minds. Table 3 lists and compares some of the key ways that these infrastructures implement FAIR and “FAIR-adjacent” practices. Following consistent design principles that support FAIR provides a level of common functionality and services that make it easier to work across these databases for an individual user or an automated agent. The more similar FAIR practices are across repositories, the more likely it is that the repositories themselves are interoperable.

A significant and positive change that is accelerating progress toward FAIR is the emergence of a set of robust standards for neuroscience data types that are starting to gain adoption. The INCF was created to help with this process of standardization and produced some early successes, e.g., the Waxholm space for registration of mouse and rat brain data (Hawrylycz et al., 2011; Papp et al., 2014), the Neuroimaging Data Model (Keator et al., 2013) the Brain Imaging Data Structure (Gorgolewski et al., 2016) were produced with support from INCF. Over the last few years, a set of standards has emerged for major neuroscience data types that can accommodate the increased size and complexity of neuroscience data through additional investments by funders and the efforts of the large brain projects, e.g., NWB, 3D-MMS (Ropelewski et al., 2022). Repositories serve as important stakeholders in ensuring that standards are followed by supporting or requiring them for data submission (Figure 2). Data uploaded to OpenNeuro, for example, must be validated against BIDS before it is accepted. The INCF has implemented an open community review and endorsement process to help improve the quality, usability, interoperability and awareness of these standards (Abrams et al., 2021). They have made available a searchable Standards and Best Practices Portfolio² where researchers can learn about each standard and how it can be used. FAIRsharing.org more broadly aggregates standards from across biomedicine and makes them available through a searchable catalog.

As neuroscience standards become more mature, better supported, and more widely used, they provide the seeds for knitting the landscape of neuroscience data repositories into a true data ecosystem, where (meta)data can flow from the laboratory to repositories and from repositories to computational tools and back again. Figure 3 shows a graph illustrating the connections between standards (light gray) and infrastructures that support them (dark gray). The data was assembled from the INCF Infrastructure Catalog, FAIRsharing, the SciCrunch Registry (Subash et al., 2023) and examination of repository websites. As shown in Figure 3, multiple repositories and infrastructures are connected via these standards. For example, the Brain Imaging Data Structure (BIDS; Gorgolewski et al., 2016) links 10 different repositories and computational platforms. The success of BIDS has led to extensions of BIDS for other modalities through a formal governance process (Governance, n.d.). The adoption of these BIDS-based standards starts to create a degree of interoperability across data types.

As tool support grows, standards are also making their way into the laboratory. BIDS, for example, has been estimated to have been used to organize over 100,000 datasets containing millions of images, indicating significant uptake by the research community (Poldrack et al., 2023). In a recent paper that outlined a neuroimaging center’s implementation of BIDS, Bush et al. (2022) stated: “Learning the BIDS specification, implementing software pipelines to map the data, and validating that the resultant mappings met the BIDS standard consumed many months of effort across multiple imaging center team members… The benefits of mapping our data to BIDS, however, far exceed the costs.” (Bush et al., 2022). These benefits included access to BID-APPS, a set of containerized analysis tools and pipelines that run on validated BIDS data, as well as improved code sharing within the lab and with colleagues, as well as a reduced barrier to publishing the data in OpenNeuro. Similarly, the electrophysiology standard, NWB, has made inroads in tackling one of the most challenging data types in neuroscience, evidenced by uptake in laboratories (Rübel et al., 2022) and support by platforms such as DANDI and EBRAINs.

Interoperability across neuroscience data has always been hampered by the multiplicity of nomenclatures and parcellation schemes from brain regions and nerve cells (Martone et al., 2004). Although slow, progress has been made. Some repositories are starting to map generic neuroanatomical structures to community ontologies like UBERON (Mungall et al., 2012). Mapping data to a common coordinate framework (CCF) allows more precise localization independent of labels applied to them (Hawrylycz et al., 2023). Encouraging signs are emerging, as CCFs for multiple species are in use or in development for the major species across the international brain projects. For example, both the BICCN/BICAN and EBrains are utilizing the Allen Institute Common Coordinate Framework v3 for mouse (Hawrylycz et al., 2009, 2023). To help manage the different versions and components that go into these atlas-based environments, a new standard for describing and versioning brain atlases was recently proposed (Kleven et al., 2023).

Standardized nomenclature for cellular taxonomies and transcriptionally defined cell types are also emerging from projects like the BICCN/BICAN to help deal with the plethora of new cell types that are emerging from new transcriptomics-based approaches (Miller et al., 2020; Tan et al., 2023). Over the years, there have been proposals for naming neurons that can bridge the multiplicity of phenotypes generated by multiple experimental techniques (Hamilton et al., 2012; Shepherd et al., 2019; Gillespie et al., 2022). However, these approaches have had difficulty in handling the complex expression patterns coming out of transcriptomics. The BICCN/BICAN recently developed the Brain Standards Data Ontology, providing a model for providing data-driven definitions of taxonomic classes (Hawrylycz et al., 2023; Tan et al., 2023). BICCN has recently introduced Cell Cards to provide a tool for exploring the BICCN taxonomic cell types for human, marmoset, and mouse primary motor cortex, including linking them to primary data sets (Hawrylycz et al., 2023). As new technologies are allowing us to derive wider scale, more complete representations of the molecular, morphological, physiological, and connectional phenotypes of neurons than was possible in the past, it is time for the global neuroscience community to come together around a common nomenclature for naming populations of cells that will aid in comparison across studies.

Services for accessing ontologies and building them into annotation and metadata pipelines have improved significantly over the past decade, with tools such as BioPortal³ and the Ontology Look Up Service⁴ providing programmatic access to community ontologies. Nevertheless, neuroscience is still a cutting edge science where many new terms are needed, particularly for annotating experimental data. For this reason, NIF and INCF had developed the NeuroLex Wiki (Larson and Martone, 2013) that lowered the barrier for creating new ontology terms. When the semantic wiki technology underlying NeuroLex was no longer available, the approach and content were ported to the Interlex on-line vocabulary management system by NIF (Surles-Zeigler et al., 2021). Interlex mints a unique identifier for each term (URI) when it is entered and allows the addition of basic metadata for each term, e.g., definition, synonyms. It also provides basic knowledge engineering functions, e.g., parent–child and other relationships, annotations. Interlex also provides various review and curation functions. These specialized terms can be used as controlled vocabularies or further engineered into ontologies as needed. Surles-Zeigler et al. (2021) provide a description of how Interlex is being used to enhance anatomical annotation of SPARC data, models and knowledge base, allowing new anatomical terms to be minted, curated, linked to existing ontologies and contributed as necessary to augment community ontologies.