Neurodata Without Borders (NWB): the open standard for neurophysiology data

Neurodata Without Borders (NWB) is an open, community-developed data standard for neurophysiology. It defines how to store recordings, stimuli, behavior and detailed metadata together in one file, most often built on HDF5, so that any lab or tool can understand and reuse the data. NWB is the standard format of the DANDI archive.

Key takeaways

• NWB solves a sharing problem: before shared standards, nearly every lab stored recordings in its own formats.
• The project was introduced in Neuron (Teeters et al., 2015) and its mature ecosystem was described in eLife (Rübel et al., 2022).
• An NWB file bundles raw data, processed data, stimuli, events and metadata with aligned timestamps.
• PyNWB and MatNWB are the main libraries, and the DANDI archive uses NWB for publishing datasets.
• Extensions let communities add new data types, which is important for emerging fields such as biocomputing.

What is Neurodata Without Borders? #

Electrophysiology produces large, varied datasets: continuous voltage traces from dozens to thousands of channels, detected spikes, sorted units, stimulus times, behavioral variables and notes about the subject. Historically each acquisition system and lab had its own formats, so reanalyzing someone else’s data meant reverse-engineering custom files.

Teeters et al. described the Neurodata Without Borders initiative in Neuron (2015) as an effort to create a common data format for neurophysiology, developed with input from multiple labs and institutes. The later eLife paper by Rübel et al. (2022), “The Neurodata Without Borders ecosystem for neurophysiological data science,” described NWB 2.x: the schema, the software stack, the extension mechanism and its adoption.

How does an NWB file work? #

An NWB file is organized like a folder tree inside a single file. The schema defines what goes where.

Every data object carries timestamps or a start time and sampling rate, so stimulation, responses and task events can be aligned precisely.

Under the hood #

NWB files are usually stored in HDF5, a format built for large hierarchical scientific datasets that supports chunking and compression. The schema itself is defined in a separate specification language, which is how NWB can be extended without breaking existing tools.

What tools make up the NWB ecosystem? #

• PyNWB: the Python library for reading and writing NWB.
• MatNWB: the MATLAB equivalent.
• NeuroConv: converters from common acquisition and processing formats into NWB.
• NWB Inspector: checks files against best practices and flags missing metadata.
• Extensions (NDX): community-defined data types for new techniques.
• DANDI archive: a BRAIN Initiative-supported repository that uses NWB for sharing and streaming datasets.

Many analysis frameworks for spikes and neural encoding and decoding can read NWB directly, which means a well-formed file plugs into existing pipelines.

What does an NWB workflow look like? #

In practice, most labs do not write NWB by hand. A common path looks like this:

1. Acquire data in the native format of the recording system.
2. Convert raw recordings into NWB with a converter such as NeuroConv, or write them directly with PyNWB.
3. Add metadata: subject or culture details, electrode layout, device description and session notes.
4. Add processed data, such as detected spikes, sorted units and trial or event tables.
5. Validate the file and run NWB Inspector to catch missing or inconsistent fields.
6. Share the file, for example by uploading it to the DANDI archive alongside a publication.

The payoff comes later. Someone who has never spoken to the original lab can open the file, see which electrode recorded which signal, when each stimulus was delivered and what the culture was, and rerun or extend the analysis.

NWB vs other neuroscience data formats #

These standards can coexist. Many labs acquire in vendor formats and convert to NWB for analysis and publication.

Why does NWB matter for biocomputing? #

Biocomputing experiments are electrophysiology experiments with extra layers: a task, a learning rule, and a living substrate whose history matters. A typical closed-loop session on a microelectrode array involves:

1. Continuous recordings across many channels.
2. Detected spikes and possibly sorted units.
3. Stimulation events, generated in real time by a closed-loop stimulation program.
4. Task variables such as a game state or input pattern.
5. Biological metadata: cell line, donor consent scope, organoid age, culture conditions.

If those pieces live in separate files with separate clocks, results are hard to verify. NWB offers one container with aligned time bases. It also connects biocomputing to the wider neuroscience tool ecosystem rather than leaving it to build everything from scratch.

Remote biocomputing platforms such as the Neuroplatform described by Jordan et al. (Frontiers in Artificial Intelligence, 2024) highlight why this matters: when many users run experiments on shared living hardware, consistent data and metadata become essential for comparison.

What still needs to be standardized #

NWB was designed largely around animal experiments, so some biocomputing details need careful handling or extensions:

• Describing organoids and cultures (differentiation protocol, age, batch) rather than animal subjects.
• Recording the logic of a closed-loop program, not only the stimuli it produced.
• Capturing consent and provenance metadata for donated cells, a topic covered in the ethics of organoid intelligence.

What are the limitations of NWB? #

• Complexity: the schema is large, and writing complete, valid files takes effort.
• Conversion work: labs must map their data into NWB structures, which can be time-consuming.
• Metadata quality: a valid file can still have thin or inaccurate metadata.
• Real-time use: NWB is mainly an archival and analysis format, not a low-latency streaming protocol for live closed loops.

Open data in NWB, with provenance and consent metadata attached, is a core part of the platform we are building at Organoid Programming. For how the underlying recordings are made, see microelectrode arrays explained, and for where data standards fit in the roadmap, read the future of biocomputing.

Frequently asked questions

What is Neurodata Without Borders?

Neurodata Without Borders (NWB) is an open data standard for neurophysiology. It specifies how to store raw and processed recordings, stimuli, behavioral data and experimental metadata together in a structured file. The goal is that any researcher or tool can open an NWB file and understand what was recorded, how and from what, without needing the original lab’s custom code.

What file format does NWB use?

NWB files are most commonly stored in HDF5, a hierarchical binary format designed for large scientific datasets. The NWB schema defines the groups, datasets and attributes inside the file. The NWB ecosystem paper (Rübel et al., eLife, 2022) describes the schema, the software that reads and writes it, and how the standard can be extended.

How do I read and write NWB files?

The two main official libraries are PyNWB for Python and MatNWB for MATLAB. Conversion tools such as NeuroConv help translate data from common acquisition systems into NWB, and NWB Inspector checks files against best practices. Many analysis tools, including spike sorting frameworks, can read NWB directly.

What is DANDI and how does it relate to NWB?

DANDI is an open archive for neurophysiology data supported by the US BRAIN Initiative. It uses NWB as its standard format for cellular neurophysiology, so datasets uploaded there can be searched, streamed and reanalyzed with common tools. Publishing data in NWB on DANDI is increasingly common when papers are released.

Why should biocomputing experiments use NWB?

Biocomputing experiments combine recordings, stimulation patterns, task events and information about the living tissue. NWB provides one place to store all of these with timestamps aligned, which makes results easier to reproduce and compare. Using a shared standard also lets biocomputing data be analyzed with the large existing set of neuroscience tools.

Sources

1. Teeters et al. (2015). Neurodata Without Borders: creating a common data format for neurophysiology. Neuron.
2. Rübel et al. (2022). The Neurodata Without Borders ecosystem for neurophysiological data science. eLife.
3. Jordan et al. (2024). Open and remotely accessible Neuroplatform for research in wetware computing. Frontiers in Artificial Intelligence.