Why use NEMAR?
NEMAR is not a replacement for every neurodata archive. It is a neurophysiology-focused access, metadata, and reuse layer that connects a human-readable dataset page, a programmatic API, a BIDS-shaped data plane, a command-line workflow, and—where conversion is available—a chunked Zarr access path.
If a dataset began somewhere else, NEMAR should make that provenance easier to follow, not hide it. The HBN EEG example below shows the difference between the source repository’s DOI and the NEMAR release DOI for the same imported dataset.
Nine reasons to use NEMAR
Section titled “Nine reasons to use NEMAR”1. One foundation, several ways to work
Section titled “1. One foundation, several ways to work”NEMAR is built around shared API, data, and metadata contracts. The website is a human-friendly shell around those contracts; the CLI and other tools use the same underlying services. You can discover a dataset in the browser, inspect it in code, and continue on a compute node without creating a second copy of the catalog or a second definition of the dataset.
See the ecosystem map and CLI versus the web.
2. Search before moving large files
Section titled “2. Search before moving large files”NEMAR exposes searchable catalog fields and dataset-level metadata before a download begins. A researcher or agent can first ask whether the modality, task, participants, sessions, or quality information look relevant, then fetch only the release that is worth analyzing.
Start with the backend API and the guide for agents and tools.
3. BIDS is the canonical source, with an explicit file contract
Section titled “3. BIDS is the canonical source, with an explicit file contract”Published datasets are exposed as BIDS-shaped trees with versioned manifests, paths, and checksums. This makes the actual release inspectable and gives scripts a stable way to resolve a file. Derived representations may improve access, but they do not replace the citable BIDS release.
Read the Data API contract for the file, manifest, metadata, and archive endpoints.
4. Neuroschema makes metadata portable downstream
Section titled “4. Neuroschema makes metadata portable downstream”NEMAR publishes dataset-level metadata as a Neuroschema
v0.4.0 document. The live on005506 metadata document
reports doc_type: dataset, source: nemar, and schema_version: 0.4.0. It gives software a
structured description of the dataset, its files, BIDS entities, authors, identifiers, terms,
and NEMAR-specific extensions. A downstream tool can use that contract instead of scraping a
page or inventing a private interpretation of every dataset.
This contract is already useful across the surrounding ecosystem. EEGDash uses NEMAR-backed records and defines typed dataset and recording schemas; its datasets can feed Braindecode workflows, while its MOABB interoperability example shows the catalog-to-benchmark handoff. MOABB also supports NEMAR as a download provider for datasets that declare a NEMAR identifier.
The libraries do not all store the raw Neuroschema document unchanged: EEGDash is an indexing and loading layer, Braindecode is a learning layer, and MOABB is a benchmarking layer. Compatibility means that the dataset identity and important metadata can travel between those layers without losing the connection to the BIDS source.
See the NEMAR metadata contract and the Neuroschema repository.
5. The DOI describes both identity and history
Section titled “5. The DOI describes both identity and history”NEMAR uses a concept DOI for the continuing dataset and a version DOI for each released state. The DOI record can carry the NEMAR identifier, creator identifiers, roles, description, domain terms, rights, funding, related publications, source relationships, and links to the operational record. The HBN comparison below shows exactly what is present in one live NEMAR/OpenNeuro pair.
6. Zarr and edge access make partial reads practical
Section titled “6. Zarr and edge access make partial reads practical”Where conversion is available, NEMAR provides a derived Zarr serving copy so a viewer or analysis client can request a time window, a few channels, or selected recordings instead of downloading an entire source file. The BIDS archive remains authoritative, and the conversion status is explicit rather than hidden behind an apparently empty viewer.
See the Zarr mental model and access recipes.
7. The CLI supports repeatable work at research scale
Section titled “7. The CLI supports repeatable work at research scale”The nemar CLI is useful when a browser is not the right tool: scripted publication, large or parallel transfers, validation in a pipeline, bulk operations, and work on a shared server or compute node. The web app remains useful for discovery and review; the two surfaces share accounts, permissions, and datasets.
8. Improvement is visible instead of silently rewriting history
Section titled “8. Improvement is visible instead of silently rewriting history”NEMAR treats a release as a fixed, citable state. A correction, new cohort, metadata enrichment, or conversion improvement can go through review, become a new version, and receive a new version DOI. The concept DOI connects the history while the version DOI protects the meaning of an older analysis.
The DOI and versioning guide explains the release workflow.
9. The interfaces are open to people, agents, and future compute
Section titled “9. The interfaces are open to people, agents, and future compute”NEMAR’s public pages, Markdown, JSON, schema.org records, manifests, API routes, and documentation are intended to be understandable to software as well as people. That supports agentic research: a tool can find a dataset, inspect its terms, identify the exact release, and explain what it used.
The compute direction builds on the SDSC and NSG partnership. Connecting NEMAR workflows to national HPC through Tapis and OneSciencePlace is a planned integration, not a claim that every dataset can already launch an HPC job. The goal is to carry the same dataset, version, provenance, and data-use constraints into a reproducible compute workflow.
A concrete comparison: HBN EEG Release 2
Section titled “A concrete comparison: HBN EEG Release 2”To make the claim that a NEMAR DOI can be more informative testable, compare the live DataCite records for one dataset rather than comparing DOI prefixes in the abstract:
- NEMAR record:
on005506, version DOI10.82901/nemar.on005506.v1.0.0, concept DOI10.82901/nemar.on005506. - OpenNeuro source record:
ds005506, version DOI10.18112/openneuro.ds005506.v1.0.1.
The NEMAR catalog identifies on005506 as imported from OpenNeuro ds005506. The two version
strings are therefore local to their repositories: NEMAR v1.0.0 and OpenNeuro v1.0.1 are not
two spellings of one universal version number. They identify releases in two different systems.
What is in each DOI record?
Section titled “What is in each DOI record?”The table describes the DataCite records checked on 2026-09-09. It compares DOI metadata, not every feature of either archive’s website.
| DataCite field | OpenNeuro source DOI | NEMAR version DOI | Why it matters |
|---|---|---|---|
| Persistent identity and landing page | 10.18112/openneuro.ds005506.v1.0.1; OpenNeuro version page | 10.82901/nemar.on005506.v1.0.0; NEMAR version page | A DOI resolves to the release-specific place where a person or tool can inspect the record. |
| Repository identifier | No separate identifiers or alternateIdentifiers in this DOI record | NEMAR: on005506 in both identifier fields | A stable local ID lets APIs, scripts, pages, manifests, and citations refer to the same NEMAR object. |
| Creators | All eight names; no ORCID or affiliation values in this DOI record | All eight names; seven ORCID iDs; no affiliation values in this particular record | Persistent person identifiers make attribution less dependent on spelling. This example is richer, but not complete. |
| Hosting and curation roles | No contributor roles in this DOI record | NEMAR as HostingInstitution and nemarAdmin as DataCurator | The record distinguishes the people who created the data from the service that hosts and curates this release. |
| Description | No description in this DOI record | Abstract plus separate citation guidance | A machine can learn what the dataset contains without first scraping a web page. |
| Subjects and keywords | No subjects in this DOI record | Domain terms including HBN, EEG, HED, CBCL, BIDS, and neuroscience | Structured terms improve discovery and give tools useful context. |
| Rights | OpenNeuro currently does not register rights metadata in its DataCite DOI records, at least in the records checked here | CC BY-SA 4.0 with an SPDX identifier and license URL | A citation record can expose the reuse condition; users must still read the license. |
| Version relationship | The DOI suffix contains v1.0.1, but the DataCite version field is empty and no version relation is recorded | version: 1.0.0, with IsVersionOf pointing to the NEMAR concept DOI | The version is explicit in metadata, not only encoded in a string. |
| Related scholarly and operational records | No related identifiers in this DOI record | IsIdenticalTo the OpenNeuro source DOI, plus the dataset paper, original HBN publications, NEMAR GitHub description, and NEMAR landing page | Related identifiers connect the imported release to its source, papers, code, and documentation. |
| Funding | No funding references in this DOI record | NIH, grant R01MH125934, and Child Mind Institute entries | Funding context is available to people and machines that build research-impact graphs. |
| File context | No sizes or formats in this DOI record | No sizes or formats in this DOI record | DOI metadata provides orientation; the NEMAR manifest remains authoritative for the exact files and checksums. |
In this specific comparison, the NEMAR record is fuller as a DataCite metadata record. That is a statement about observable field coverage, not a claim that NEMAR is intrinsically a better DOI, that OpenNeuro’s website lacks the information, or that every NEMAR record is equally complete. Both DOIs are valid scholarly identifiers. The difference is what each repository chose to put into its registrar record.
For example, OpenNeuro’s own documentation describes its BIDS requirement and public-domain licensing policy even though those details are not registered in the current OpenNeuro DataCite records checked here. The NEMAR record states CC BY-SA 4.0 for the NEMAR release. Those are separate repository statements: read the license attached to the release you actually use rather than assuming that an imported dataset has the same terms everywhere.
Why NEMAR mints its own DOI
Section titled “Why NEMAR mints its own DOI”The DOI prefix is not a quality score. NEMAR uses its own DOI namespace so that the NEMAR release has a persistent scholarly identity in the NEMAR system:
10.82901/nemar.on005506identifies the continuing NEMAR dataset concept.10.82901/nemar.on005506.v1.0.0identifies one released NEMAR state.on005506connects that identity to the NEMAR API, data plane, GitHub metadata repository, dataset page, and derived access services.
This lets NEMAR improve the resource without silently changing what an old citation means. A correction, added cohort, metadata enrichment, or conversion improvement can become a new reviewable release. The concept DOI connects the history; the version DOI tells another researcher which state an analysis used.
DataCite’s own guidance describes this pattern as a canonical DOI connected to version-specific
DOIs with HasVersion and IsVersionOf relations. NEMAR uses that relationship for its concept
and version records. See DOI and versioning for the contributor
workflow and citation rules.
When to cite which DOI
Section titled “When to cite which DOI”Use the DOI for the resource you actually used:
| Your workflow | Citation choice |
|---|---|
| You downloaded the original OpenNeuro snapshot directly | Cite the OpenNeuro version DOI. |
You used the NEMAR BIDS release from data.nemar.org or the NEMAR CLI | Cite the NEMAR version DOI. |
| You used NEMAR’s release and want to preserve the source provenance | Cite the NEMAR version DOI and the OpenNeuro source DOI. |
| You are discussing the NEMAR dataset across releases | Use the NEMAR concept DOI, and name the analyzed version when it matters. |
For any reproducible analysis, save the dataset ID, version DOI, manifest or snapshot information, analysis code, and relevant parameters. A DOI tells readers which scholarly object to find; the version and manifest tell them which files and metadata state the analysis consumed.
The honest boundary
Section titled “The honest boundary”NEMAR and OpenNeuro can serve complementary purposes. OpenNeuro remains the source archive for datasets deposited there, with its own BIDS requirements, access mechanisms, and version DOIs. NEMAR is the better fit when you want the NEMAR catalog, neurophysiology-oriented enrichment, NEMAR’s API/CLI/data-plane contracts, citable NEMAR releases, or derived partial-read access.
The right question is not “which DOI looks more official?” It is “which release, metadata record, and access path did my work actually use?” NEMAR’s goal is to make that answer explicit and easy to reproduce.