This page provides a compact overview of the Triple-N dataset, focusing on how the data are organized, which files are most important for analysis, and what users should download first.

Main data components

The current ScienceDB release is organized into three top-level folders: Processed, Raw, and others.

Processed

Processed contains compact session-level summary files, with one file per recording session. These files are designed as analysis-oriented outputs rather than full raw recordings. In general, this part of the release stores per-unit summary measures and response summaries that are convenient for downstream statistical analyses, selectivity analyses, and quick session-level screening.

Raw

Raw contains the main session-wise neural data products and their higher-dimensional exports. In the current release, it is further divided into several subfolders:

  • GoodUnit - Session-wise unit files that store the main spike-aligned neural response data together with unit properties and session metadata. This is the most complete per-unit representation for many downstream analyses.
  • GoodUnitRaw - Session-wise unit files that preserve a less reduced version of the unit-level outputs. These files are useful when users want richer intermediate data rather than only compact summaries.
  • GoodLFP - Session-wise local field potential files. This part of the release is intended for analyses of trial-aligned LFP responses and associated recording metadata.
  • H5FILES - HDF5-based session exports for efficient loading of large response arrays. These files are paired with companion metadata files and provide a faster, more array-oriented access path than the MATLAB struct-based files.
  • SessionFolder - Session-wise original data, directly copied from the recording system, serve as input to the preprocessing pipeline. Currently, each session folder is compressed as a .tar.gz archive and can be downloaded from the raw section of the dataset link.

Overall, Raw is still organized by session, but within each session it separates spike-based data, LFP-based data, and fast-loading array exports into different file groups.

others

others contains supporting resources that are shared across sessions rather than tied to a single recording. In general, this folder includes auxiliary metadata tables, image- or stimulus-related resources, model-feature packages, and other companion files that help interpret or align the neural data. It is the main place to look for cross-session annotations and supplementary materials rather than the core session-wise neural response files.

Triple-N is best downloaded based on analysis goals rather than as a complete archive, as many users do not need every component of the release. For most use cases, the Processed and Raw/GoodUnit directories provide the lightest entry point and are typically sufficient. These include compact session-level outputs, per-unit summary measures, detailed spike-aligned responses, unit properties, and session metadata.

For large-scale workflows where loading speed is critical, Raw/H5FILES offers a more efficient, array-based format and is often the preferred choice.

The SessionFolder-level data are substantially larger (~3 TB) and are primarily intended for researchers who wish to re-run spike sorting, refine eye fixation detection, or recompute quality metrics.