Glossary#
249 terms across 19 categories, from the Allen Glossary. Search matches names, definitions, categories and dataset names; the category legend doubles as a filter, so clicking one or more pills narrows the list. Every term has a permalink you can paste into an email — click a term name to copy the link to it.
Category — the colour on a card's edge. Click to filter.
Illustration — colour inside a drawing means anatomy, never category
The illustrations are generated rather than hand-drawn. They are being checked by the people who know the data, but errors cannot be ruled out at this stage — read them as sketches of the idea, and trust the definition over the picture. 148 of 249 terms have one.
3D reconstruction
Turning EM imagery into 3D neuron objects (dense segmentation → meshes).
Action potential
A characteristic signal in excitable cell membranes: a potential-difference waveform that propagates along the membrane. In neurons it indicates activation. The trace is a Hodgkin-Huxley simulation: a brief current pulse opens sodium channels, which depolarise the membrane and then inactivate, while potassium conductance rises more slowly and repolarises it past rest.
AIND metadata schema
Six JSON classes describing a newer data asset: data description, subject, procedures, rig or instrument, session or acquisition, and processing. Where you look up which virus was injected, or what a capsule actually ran.
AllenSDK
The Python package for the Brain Observatory physiology datasets, wrapping downloads and metadata behind a cache object. Being retired in favour of reading NWB files directly, so new work should not start here.
amplitude_cutoff
Estimated fraction of the unit's spikes that fell below the detection threshold and were never recorded — a false-negative rate. Default threshold 0.1.
Annotation
Labeled data (points/tables) bound to locations or cells in the volume.
Baiting / coupled vs uncoupled
Baiting: a reward an unchosen side would have given is held and delivered on the next choice of that side. Coupled or uncoupled describes whether the two sides' probabilities change together or independently.
Basket cell (BC)
Inhibitory neuron whose synaptic output targets the cell body and proximal dendrites of excitatory neurons. Many basket cells express parvalbumin (PV), but not all — some express cholecystokinin (CCK). PV basket cells are typically fast spiking, and are thought to be important for gain control and for the temporal precision of network activity.
BCI task / conditioned neuron
A lickport moves toward the mouse at a speed set by the fluorescence of one chosen neuron. Reaching the near position within 10 s earns water. Mice usually learn to drive that neuron within about 30 trials.
Behavior session
One behavioural recording, whether it happened under the microscope or in the training facility. Its session_type names the training stage, which is how the full training history is reconstructed.
Bipolar cell (BPC)
A subset of VIP cell with a bipolar dendritic arbor — two primary dendrites leaving opposite poles of the soma. Distinct from the retinal cell of the same name.
Blank sweep
A trial in which the stimulus is replaced by mean-luminance grey, interleaved among real trials so each stimulus has its own baseline.
Bound Spatial Point
Binds an annotation to the cell at a location via the triad pt_position → pt_supervoxel_id → pt_root_id.
Branch / End / Root point
Named skeleton vertex types; the root is conventionally placed at the soma.
Catch trial / sham change
A change time is drawn but the image does not change. This conservative definition counts only presentations drawn from the change-time distribution; aborted trials are arguably catches too.
CAVE
Connectome Annotation Versioning Engine — the suite managing large dynamic connectomics data.
CAVEclient
The main Python client for programmatic access to CAVE services. Servers: MICrONS global.daf-apis.com, V1DD global.em.brain.allentech.org.
cc_abs / cc_max / cc_norm
Digital-twin model-performance columns.
Cell type
Classification of a cell (e.g. 23P, BC) via several tables/methods, keyed on nucleus id.
cell_id / soma_id
The 6-digit nucleus id (from nucleus_detection_v0), static across versions; tracks a cell over time.
cell_specimen_id vs cell_roi_id
cell_roi_id identifies a segmented ROI within one experiment, before matching. cell_specimen_id identifies the cell after matching across sessions, and is therefore shared across a container. Joining on the wrong one silently loses the across-day link.
Change detection task
A go/no-go task: images are presented in a continuous stream and the mouse earns water by licking when the image identity changes. The 500 ms grey gap between images adds a working-memory component.
Channelrhodopsin (ChR2)
A light-gated ion channel used in optogenetics to control neuronal activity with light.
Channels table
One row per recording site, at general/extracellular_ephys/electrodes, with its position on the shank and in the CCF. A unit points into this table through its electrodes column; that is how a spike acquires a place in the brain.
ChRmine
A red-shifted opsin, excited near 1080 nm. Because GCaMP is excited near 920 nm the two can be driven independently, which is what makes simultaneous imaging and single-cell photostimulation possible.
classification_system column
The E / I / non-neuron grouping column in cell-type tables.
Clean
Arbor proofread to remove all merge errors (synapses correct, but may be incomplete).
cloud-volume / ImageryClient
Serverless clients to read Precomputed imagery/segmentation and download aligned cutouts.
Column (MICrONS)
A 100 µm-square region spanning all cortical layers, densely proofread for a cell-type census.
Column (V1DD field)
A column field naming one of 5 stacked scan sub-volumes tiling the V1DD block — a different concept from the MICrONS column.
Common Coordinate Framework (CCF)
A standard 3D reference space for the mouse brain that lets data from different modalities be placed in the same coordinates.
Compartment labels
SWC integer codes: 0 undefined, 1 soma, 2 axon, 3 basal dendrite, 4 apical dendrite.
Connectivity Viewer
Dash app showing a cell's synaptic inputs/outputs grouped and colored by cell type.
Connectome
A wiring map of neurons and the synaptic connections between them.
Container
There is no consistent use of this term.
Context block
A ten-minute stretch in which only one modality is rewarded, signalled by instruction trials at its start. Blocks alternate for six blocks in a session.
Coordinate frames
Three systems: voxel (annotations), nanometer (mesh/skeleton vertices), transformed (pia-flattened microns).
Coregistration
Aligning functionally-imaged cells to the same cells in the EM volume (manual + automatic).
Cre line
Cre recombinase catalyses recombination between loxP sites. Paired with a loxP reporter line it drives the reporter's expression, and because Cre is expressed within a specific gene the expression is restricted to a subset of cells.
ctr_pt_position
The synapse-junction center point (not root-id-bound).
Current source density (CSD)
The second spatial derivative of the LFP along the probe, which localises current sinks and sources and so the laminar position of synaptic input.
d_prime (unit)
Separability of this unit's waveforms from its neighbours', by linear discriminant analysis. Higher is better. Not the behavioural d-prime.
d-prime (behavioural)
Signal-detection sensitivity for the task: how far the hit rate exceeds the false-alarm rate. Not the unit quality metric of the same name.
Dash web apps
Plotly-Dash apps (Table Viewer, Connectivity Viewer) for fast querying + Neuroglancer-link generation.
Datastack
A named bundle of imagery + segmentation + annotation DB (minnie65_public, v1dd_public).
decoder_label
The pipeline's automated call on what a unit is — sua for a single unit, and so on — with decoder_probability as its confidence.
Default quality filtering
Visual Coding applies isi_violations, amplitude_cutoff and presence_ratio filters by default; Visual Behavior Neuropixels returns every unit unfiltered. Same SDK, opposite defaults — check which you are holding.
default_qc
A single pass/fail flag summarising the pipeline's quality criteria for a unit, in the AIND-packaged datasets.
Depth / pia→WM axis
y increases with cortical depth, so depth plots need ax.invert_yaxis().
Digital twin
A DNN trained to predict a cell's response to arbitrary stimuli (source of derived functional properties).
Direct vs indirect activation
The central pitfall of optotagging: a neuron may respond to the laser because it expresses the opsin, or because a neuron that does synapses onto it. Direct responses are short-latency (<10 ms), reliable across pulses, and tightly distributed in time.
Distance
Four geometric senses and two statistical ones are in routine use, and they give different answers for the same pair of points.
Drift metrics
max_drift and cumulative_drift record how far, in µm, a unit's spikes moved along the probe during the session. Newer pipelines add activity_drift and drift_ptp.
Drifting gratings
A full-field sinusoidal grating moving orthogonal to its own orientation. Parameters: orientation and direction (degrees), temporal frequency (Hz), spatial frequency (cycles/deg), contrast. Typically 2 s on, 1 s grey.
Driver line
A transgenic line engineered to label a specific cell population by expressing a gene under that population's promoter. The driver line determines which cells are targeted; the reporter line determines what is expressed in them.
DSI
Direction selectivity index (0–1).
Dynamic foraging task
Two choices, binary reward, and reward probabilities that change during the session. A go cue opens a short window in which the mouse licks left or right; the mouse must learn from recent outcomes to track the better side.
Dynamic Routing task
A context-dependent go/no-go task alternating visual and auditory blocks. The same stimulus is a target or not depending on the current block, so stimulus and meaning can be separated.
Edges
Pairs of connected vertices (mesh.edges, skeleton edges).
Electron microscopy (EM)
Imaging that reaches nanometer resolution to reveal tissue ultrastructure.
Encoding vs decoding
Encoding asks whether an event changes neural activity; decoding asks whether the event can be read back out of the activity. Same data, opposite direction.
Enhancer AAV
A virus carrying a cell-type-specific enhancer, used to restrict expression without breeding a transgenic line.
Environment secrets
How the CAVE auth token is supplied when code runs on a shared or hosted machine: exported as environment variables named API_SECRET_<server> instead of being written to a credentials file in the home directory.
Ephys
Shorthand for electrophysiology.
Ephys selection bias
Spike sorting needs enough spikes to form a cluster, so sparsely active neurons are missed and large-spike, high-rate neurons — and layer 5 — are over-represented. Ophys sees many of the cells ephys does not.
Epoch
A labelled stretch of time — but of what, and on whose clock, differs everywhere it appears.
Error profiles
The characteristic ways automated segmentation fails, and how they differ by compartment: thin axons are dominated by split errors, thicker dendrites and somata by merges. This asymmetry is why proofreading status is tracked separately for axon and dendrite.
Event detection
Deconvolving ΔF/F into discrete events, here with the L0 method. At population imaging resolutions 1- and 2-spike events are detected unreliably, particularly with GCaMP6f.
Evoked vs spontaneous
Activity driven by a stimulus versus activity during the grey-screen epochs. The comparison that decides whether a response is a response at all.
Excitatory V1 cell types
Pyramidal subclasses by layer/projection: 23P, 4P, 5P-IT/ET/NP, 6P-IT/CT (+ mtype clusters L2a…L6wm).
Experience level
Whether the image set in a session is the one the mouse trained on (Familiar) or a different one (Novel). The axis the Visual Behavior datasets were built to test.
Experiment
There is no consistent use of this term. Establish which one is meant before joining anything.
Extended
Arbor proofread to remove all merge AND split errors (correct and as-complete-as-possible).
Extracellular electrophysiology
Recording voltage from outside the cell membrane, which gives better access to intact brains than intracellular recording. Its two readouts are spikes and the local field potential.
Eye tracking / pupil
Ellipse fits to eye, pupil and corneal reflection per video frame, giving area, centre and rotation, plus a likely_blink flag. Recorded during physiology sessions but not during training.
Faces
Triangles of connected vertex indices that tile a mesh surface (mesh.faces).
Fast spiking neuron (FSN)
Narrow, fast action potentials; with enough injected current, high spike rates without frequency adaptation. In unlabelled extracellular recordings, narrow-waveform units are called fast spiking and putatively identified as PV+ cells.
FIBSEM
Focused-ion-beam SEM; block-face EM that mills & images, giving near-isotropic voxels.
Field of view
The imaged extent of one plane, in pixels and in µm. Recorded per experiment as field_of_view_width/height.
firing_rate
Mean spike rate over the whole session. Low values may mean a sparsely active neuron or a badly detected one.
Fluorophore
A molecule that absorbs light and re-emits it at a longer wavelength. Fluorophores fluoresce only while exposed to a light source.
Functional connectome
A dataset linking synapse-resolution EM connectivity to recorded neural function in the same neurons.
GABA
The main inhibitory neurotransmitter in the mammalian brain. In cortex most GABAergic neurons are local interneurons.
Gabor patches
Spatially restricted gratings. The receptive-field mapping stimulus in Visual Coding Neuropixels: 20° diameter, three orientations on a 9 × 9 grid of screen positions, identical in every session.
GCaMP
A family of GECI fusing calmodulin's calcium-binding domain to green fluorescent protein. GCaMP6f and 6s are the fast and slow variants, differing in sensitivity and especially in decay kinetics.
Genetically-encoded calcium indicator (GECI)
A protein expressed by a cell that changes its fluorescence on binding Ca²⁺, used to visualise neural activity with fluorescence microscopy.
Golden Mouse (409828)
The single V1DD mouse with functional coregistration.
gOSI / gDSI
Global orientation/direction selectivity indices (vector-sum variant).
Graphene (graphene://)
URL protocol for dynamic, CAVE-backed (editable) segmentation/meshes, vs static precomputed://.
Graphene vs Precomputed
graphene:// = dynamic/editable; precomputed:// = static.
Grids / Chunk
The volume is partitioned into a 3D grid of chunks for the chunked-graph.
Head fixation / head bar
A surgically implanted bar clamps the mouse's head in a repeatable position — better than 10 µm across clamp cycles, which is what makes it possible to return to the same cells on a later day.
Higher visual area (HVA)
A cortical visual area receiving input from primary visual cortex, and so higher in the visual hierarchy. In the mouse: VISl, VISal, VISpm, VISam, VISrl among others.
Hit / miss / false alarm / correct reject
Lick within the 750 ms window after a change = hit; no lick after a change = miss; lick after a sham change = false alarm; withholding on a sham change = correct reject. Licking before the scheduled change aborts the trial.
Image set
Which eight natural images a session used (G or H, A or B). Two images are shared between sets, so novelty is a property of the other six.
Imagery
The 3D grayscale (0–255) array depicting EM ultrastructure.
Imaging depth
Depth in µm below the cortical surface at which a plane was collected. Roughly: <250 layer 2/3, 250–350 layer 4, 350–500 layer 5, >500 layer 6 — but layer-specific Cre lines are the reliable way to get layer specificity.
Imaging plane
One two-photon focal plane. A single-plane microscope images one per session; the Multiscope/Mesoscope images up to eight. The plane, not the session, is what an ophys experiment is defined on.
Indicator sparsification
Calcium indicators respond non-linearly to firing rate: bursts are boosted, isolated spikes washed out. Tuning measured with ophys therefore looks sharper and sparser than the same tuning measured with ephys.
Inhibitory V1 cell types
Interneuron subclasses: BC, BPC, MC, NGC (manual) and PTC/DTC/STC/ITC (targeting-based mtypes).
Interneuron
A neuron with short axons that synapses only with nearby neurons. In cortex the term is often used to mean an inhibitory neuron.
Intrinsic signal imaging (ISI)
Measuring blood-flow changes from the reflectance of red light on the brain surface. Commonly used to map retinotopy across the cortical surface and so to target later recordings.
isi_violations
Rate of inter-spike intervals shorter than the refractory period. A real neuron cannot fire that fast, so violations mean spikes from more than one cell were merged. Default threshold 0.5.
isolation_distance
Distance in Mahalanobis space to the nearest other cluster of waveforms. Higher is better separated.
IT / ET / NP / CT / SP
Projection categories: intratelencephalic, extratelencephalic, near-projecting, corticothalamic, subplate.
Kilosort
The template-matching sorter used for all Allen Neuropixels data. It merges automatically, so no manual curation step is needed for recordings with little drift.
l_ratio
Contamination measure related to isolation distance: the probability that nearby spikes belong to this cluster. Lower is better.
Layer (cortical)
L1–L6 along the pia→WM axis; drives cell-type naming. NOT the Neuroglancer layer.
Level of detail (LOD)
Static meshes are smaller, multi-LOD, precomputed://; dynamic meshes are detailed, single-LOD, graphene://.
Local field potential (LFP)
Transient electrical potential generated in nervous tissue by the summed activity of the cells in it, typically measured below 250 Hz. Informative about oscillations and network synchrony.
Locally sparse noise
Black and white spots flashed on a grey screen, arranged so no two spots fall within 5 pixels of each other. The exclusion zone is what makes the average around any pixel structureless, so a receptive field can be recovered.
Manifest
The file a cache uses to know what data exists and where it was put. Instantiating a cache without naming one creates it in the working directory. There is no manifest when you read NWB directly; the file is the manifest.
Martinotti cell (MC)
A subtype of SST cell that targets the apical dendrites of pyramidal cells in layer 1. Martinotti cells are found in layer 2/3 and layer 5.
Materialization & Versioning
Timestamped snapshots of the annotation DB; each version = a fixed timestamp (MICrONS v1507, V1DD v1196).
Maximum / average projection
The imaging movie collapsed over time into one image — the standard way to see every cell in a plane at once.
Merge errors
Two neurons' processes incorrectly joined; they add false connections.
Meshes
Vertices + triangular faces defining a neuron's 3D outer surface.
MeshParty / Meshwork
Python package + object bundling the L2 mesh, skeleton, and anno annotations, kept in sync.
Meshpoints
Informal usage for mesh vertices. Not a formal term — say vertices, since “point” elsewhere means an annotation position.
MICrONS
Cubic-millimeter functional-connectomics EM dataset of mouse visual cortex (VISp/VISal/VISrl).
Minnie
Internal name for the MICrONS dataset/mouse (minnie65; datastack minnie65_public).
Motion correction
Registering every frame of the imaging movie to a reference before segmentation, so an ROI mask refers to the same cell throughout.
mtypes
Morphology/connectivity-derived cell-type clusters (L2a…L6wm; PTC/DTC/STC/ITC).
Natural movies
Black and white film clips with natural spatial and temporal statistics — usually the opening shot of Touch of Evil, chosen because it is continuous, with no cuts and varied motion.
Natural scenes
Black and white photographs with natural spatial statistics, flashed for 0.25 s with no gap. Visual Coding uses 118 images drawn from the Berkeley, van Hateren and McGill image sets.
Neuroglancer
WebGL browser viewer for very large volumetric connectomics data (imagery, segmentation, meshes, annotations).
Neuroglancer forks
Neuroglancer is maintained as several diverging branches. Spelunker is the one CAVE datastacks link to; the Seung-lab and FlyWire branches are the other widely used ones. States are broadly compatible but not identical.
Neuroglancer Layer (img/seg/ann)
The data layers in a Neuroglancer state. NOT the cortical layer.
Neuroglancer State
JSON object storing all layers/view/annotations, identified by a state id.
Neurogliaform cell (NGC)
An interneuron that makes a diffuse axonal arbor and is thought to release GABA through both synaptic release and volume transmission, non-selectively inhibiting nearby neurons.
Neuronal process
An axon or dendrite branch of a neuron (a process that splits at branch points).
Neuropil correction
An annulus around the ROI, excluding nearby cells, gives a local neuropil signal. It is subtracted from the raw trace after weighting by a per-cell r value.
Neuropixels
A family of silicon probes for high-channel-count single-unit extracellular recording, miniaturised with integrated-circuit design so that hundreds of units can be recorded from one probe with minimal brain damage.
nglui (statebuilder/parser)
Python package to generate and parse Neuroglancer states from dataframes.
nn_hit_rate / nn_miss_rate
Nearest-neighbour estimates of contamination and of missing spikes respectively.
Nodes
Vertices in the skeleton / L2 graph.
NP 1.0 / 2.0 / Ultra / Opto
1.0: 960 sites, ~20 µm pitch, ~3.8 mm span. 2.0: 1280 sites per shank, ~15 µm pitch. Ultra: 6 µm pitch, fine detail over a shorter span. Opto: 1.0 plus 28 on-shank light emission sites. All read out 384 channels at a time.
NWB (Neurodata Without Borders)
The standard file format for physiology and behaviour data. Visual Coding and Visual Behavior use an HDF5 backend; the newer datasets — V1DD, BCI, Dynamic Foraging, NP Ultra — use a Zarr backend optimised for cloud access.
NWB layout
Every NWB file has the same top-level groups: general (subject, devices, electrodes or imaging planes), acquisition (signals as acquired), stimulus (what was presented), intervals (epochs, trials, blocks), processing (anything derived), units (sorted units, ephys only) and analysis (non-standard extras). What differs between datasets is what fills them — and where a dataset puts a thing is not always where you would guess, so print the tree first.
Omission
5% of non-change presentations are dropped, interrupting the expected stimulus cadence so that expectation signals can be measured. Omissions occur during recording but not during training, and never at or just before a change.
Ophys
Shorthand for optical physiology, often in reference to two-photon calcium imaging, but can also include other methods such as fiber photometry.
Ophys container
The same imaging plane followed across days. Containers hold different numbers of sessions depending on which passed QC and how many retakes happened.
Ophys experiment
One imaging plane in one session — the narrowest unit in the hierarchy, with its own imaging_depth and targeted_structure. Quality control passes or fails each plane separately.
Ophys session
One continuous recording under the two-photon microscope. It contains one imaging plane on a single-plane scope and up to eight on the Multiscope.
Opsin
A light-gated ion channel. Illumination changes its conformation, letting ions cross the membrane and either forcing the cell to spike (excitatory opsin) or suppressing spiking (inhibitory).
Optogenetics
Controlling neural activity by expressing light-activated ion channels in a specific subpopulation — a reporter line for the opsin, a driver line for the population — giving temporally precise control of spiking.
Optotagging
Using optogenetics to identify which recorded units belong to a genetically defined population, by their response to laser pulses. Trains of 10 ms pulses at 20 Hz are a common stimulus.
Oracle score
Visual-response reliability — signal correlation across repeated “oracle” movies.
OSI
Orientation selectivity index (0–1).
Parvalbumin-positive (PV+) neuron
Fast-spiking GABAergic interneurons with strong inhibitory effects on their neighbours; action potentials can be under 400 µs. Parvalbumin is a calcium buffer, so calcium imaging of these cells should be read cautiously.
Passive replay block
The same stimuli replayed with the lick spout retracted and no reward, so task-dependent modulation can be separated from stimulus drive.
Peak channel
The channel on which a unit's mean waveform is largest. A unit carries no position of its own — joining peak_channel_id to the channels table is how it acquires a CCF location, a brain-region label and a depth.
Physiology
The activity side of a functional-connectomics dataset: the calcium-imaging responses recorded from the same neurons that were later reconstructed in EM.
Position
The 3D coordinate of a bound spatial point (pt_position, stored in voxels by default).
Precomputed format
Storage representation for arbitrarily large images/meshes/skeletons.
pref_dir
Preferred direction in degrees (0–360; 0 = vertical bar moving right, CCW+).
pref_ori
Preferred orientation in degrees (0–180).
presence_ratio
Fraction of the session in which the unit had spikes. A low value usually means the unit drifted away from the probe. Default threshold 0.9.
Probe / shank / channel / site
The recording hierarchy: a probe carries one or more shanks, a shank is patterned with recording sites, and the subset wired out for recording at any moment are the channels.
Project cache
The AllenSDK entry point for the Brain Observatory datasets: it downloads what you ask for, keeps it in a known directory, and hands back manifest tables and session objects. Newer datasets have no cache — you open the NWB file yourself.
Proofreading
Manual correction of split/merge errors to make neurons biologically accurate/complete.
PSTH
Peri-stimulus time histogram: spikes binned relative to stimulus onset and averaged over trials, giving the time course of the response.
PyChunkedGraph (PCG) / L2 graph
Hierarchical representation: L0 = voxels, L1 = supervoxels, L2 = supervoxels grouped within a chunk.
Pyramidal cell
An excitatory neuron with a characteristic cell-body shape and apical dendrite. In visual cortex, by far the most common excitatory type.
Q value / RPE
Latent variables of a reinforcement-learning fit to foraging behaviour: the expected value of each choice, and the reward prediction error that updates it. Useful precisely because they can then be regressed against neural activity.
query_table / synapse_query
The two query entry points + filter_in_dict; note the 200k-row cap, desired_resolution, select_columns, split_positions.
Radius
Half the cable thickness at a skeleton vertex (µm).
readout_loc_x/y
Approximate receptive-field center in stimulus space.
Receptive field
The region of the stimulus domain in which a stimulus must lie to evoke a response. Generalises beyond space to any stimulus dimension, and so to the stimulus features that drive a cell.
Reference table
A table linked to another (usually nucleus_detection_v0) by shared annotation id, adding _ref columns.
Regular spiking neuron (RS)
Longer action potentials and spike-frequency adaptation — the rate falls over a sustained current step. The most common cortical type, usually associated with excitatory pyramidal neurons.
Reporter line
A transgenic line engineered to express a protein that monitors or manipulates activity — GFP, GCaMP, channelrhodopsin — but only once the controlling protein (Cre or FLP) is present.
Residual / Separation score
The two coregistration-quality metrics.
Resolution
Physical voxel size in nm/voxel (MICrONS 4×4×40; V1DD 9×9×45); set per query via desired_resolution.
Response modulation index (RMI)
The normalised contrast between visual and auditory target response rates, collapsing two hit rates into one number that says which context the mouse is behaving in.
Retake
A second attempt at a session_type after the first failed QC. Why prior_exposures_to_image_set and not session_type tells you whether a session was truly the first with novel images.
Retinotopy
The mapping of visual space onto neural space: neighbouring points in the visual field are represented by neighbouring points in the brain. Measured as altitude (upper–lower) and azimuth (left–right).
ROI mask
The pixel mask for one segmented cell in an imaging plane. In two-photon data an ROI is the set of pixels thought to belong to a single neuron.
Root_id (pt_root_id)
Unique integer for a specific segmentation = a specific version of a cell (a.k.a. segment / object id).
Running speed
Speed on the running disc, temporally aligned to the activity traces. Same length as ΔF/F, so a stimulus epoch indexes into both.
Saccade
A rapid ballistic eye movement between fixation points. Mice are not foveal animals and their eye movements differ from those of foveal species.
Scan
The scan_idx from functional imaging; part of the ROI's unique id.
Segmentation
A 3D array where each voxel stores the root_id of the object at that location.
Segments (= root/object id)
“Segment id” used as a synonym for root id — collides with the skeleton sense of “segment”.
Segments (skeleton)
An unbranched run of vertices between branch/end points.
Serial-section EM
Many ultrathin sections are cut from a block, imaged one by one, then re-aligned into a volume. Resolution is fine in x/y and coarse in z, so voxels are strongly anisotropic.
Session
The databook defines it as “a physiological and/or behavioral recording that happens at one time”, but four narrower senses are in use as identifiers.
Share link / middleauth
Authenticated state-sharing mechanism.
Signal vs noise correlation
Signal correlation compares two cells' mean responses across stimulus conditions — do they like the same things. Noise correlation compares their trial-to-trial fluctuations to the same condition — do they vary together.
Single unit vs multi-unit
Not two categories but a gradient, from complete and uncontaminated to incomplete and highly contaminated. Every analysis still has to draw a binary line somewhere; quality metrics are how you draw it deliberately.
Skeletons
Tree-like linear representation of a neuron's branching (vertices + edges, radius, compartments).
snr
Waveform amplitude relative to background noise on the peak channel.
Somatostatin (SST) cell
An inhibitory interneuron expressing somatostatin (SST, sometimes SOM). SST cells tend to target the distal dendrites of excitatory neurons, and have important roles in regulating their activity.
Source
Disambiguation: image_source/segmentation_source, the Neuroglancer layer source, and skeleton path_between(source,…).
Source (presynaptic)
The presynaptic partner of a synapse (pre_pt_root_id).
Spatial frequency
How often the sinusoidal components of a signal repeat per unit distance — for a grating, the spacing of its bars. Typically cycles per degree.
Spike band / LFP band
The two streams split off each channel: the spike band at 30 kHz with a 500 Hz high-pass, carrying action potentials from adjacent neurons; the LFP band at 2.5 kHz, carrying low-frequency fluctuations from a wider area.
Spike raster
One row per trial, one tick per spike, aligned on an event. The plot to make before any model, because it shows trial-to-trial structure that an average hides.
Spike sorting
Assigning detected spikes to individual neurons — a blind source separation problem. Detection, extraction, feature extraction, clustering, then validation against the refractory period.
Split errors
A process incorrectly appears to stop; they remove true connections.
Spontaneous activity
An epoch of mean-luminance grey with no patterned stimulus, included in most sessions as a baseline for visually evoked activity.
standard_transform
Package converting voxel/nm coordinates to pia-flattened micron coordinates (minnie_ds, v1dd_ds).
State
Four unrelated meanings, two of which appear in the same workshop.
Static gratings
A stationary full-field sinusoidal grating flashed for 0.25 s. No temporal frequency; phase becomes a parameter instead.
Status flags
Booleans status_axon/status_dendrite recording whether each arbor was proofread, plus valid_id (root id at assessment).
Stimulus epoch table
When each interleaved stimulus block began and ended. In Visual Coding 2P the bounds are given as imaging frames, so they index directly into the ΔF/F and running-speed traces.
Stimulus presentations table
One row per stimulus shown, with its parameters and its start_time and stop_time. The table every alignment starts from. In NWB it lives under stimulus/presentation, or as a TimeIntervals table under intervals — which one depends on the dataset.
Stimulus template
The literal image shown, stored alongside the stimulus table for image and movie stimuli. Often available both unwarped and warped — the warped version is what the monitor rendered.
Strategy values
dendrite_clean, dendrite_extended, axon_partially_extended, axon_fully_extended, axon_interareal (MICrONS only), none.
Structure acronym
The CCF region label attached to a channel or unit — VISp, MOs, LSr. A unit with no CCF registration gets coordinates of [-1, -1, -1].
Supervoxel (pt_supervoxel_id)
L1 grouping of voxels within a chunk; the stable internal id an annotation binds to.
Surround suppression
A stimulus extending beyond a cell's classical receptive field suppresses its response. Stronger in superficial layers, and one of the questions V1DD's windowed and full-field gratings were designed to address.
SWC format
Standard skeleton file format (one of three: SWC, meshwork-h5, precomputed).
Synapse size
Synapse size in voxels; correlates with surface area / strength.
synapse_target_predictions_ssa
Per-synapse postsynaptic-compartment prediction (soma / spine / shaft).
synapses_pni_2 / synapses_v1dd
The sole synapse tables (337M / 639M rows).
Table Viewer
Dash app to query/filter one table and select rows in Neuroglancer.
Tables
CAVE annotation tables (synapses, nuclei, cell types, proofreading, coregistration).
Tags / Shortcuts
Keyboard-driven annotation labels for fast bulk labeling in Neuroglancer.
Target
Disambiguation: target_id (reference link) vs synaptic postsynaptic partner vs path target_index.
Target (postsynaptic)
The postsynaptic partner of a synapse (post_pt_root_id).
TEASAR
Algorithm that turns the L2 graph into a skeleton tree.
TEM
Transmission EM; MICrONS/V1DD are serial-section TEM-style (thin sections, anisotropic z).
Temporal frequency
How many complete periods the signal goes through per unit time. Typically Hz.
Three-photon (3P) imaging
Raises signal-to-noise for deep imaging of densely labelled tissue. Used to extend the V1DD centre column to white matter, where 2P image quality has degraded.
Token / auth
Google-account credential required before any programmatic access, saved per server.
Transgenic line
A mouse line whose genome has been altered by introducing foreign DNA. Here, typically a Cre line driving expression of a reporter line within a specific subset of cells.
Trials table
One row per trial: timing landmarks and outcome flags. Usually nwb.intervals['trials'], but not always — the BCI dataset keeps its trials under stimulus/presentation, because there the lickport is driven by the neuron. And a “trial” is not always behavioural: in the cell-type look-up table it is a laser pulse train.
Tuning curve
Mean response plotted against a stimulus parameter. The shape of the curve is what selectivity indices such as OSI and DSI summarise in one number.
Two-photon calcium imaging
Measuring neural activity through a fluorescent calcium indicator such as GCaMP. At rest a neuron has low calcium; when it spikes, calcium flows in, binds the indicator and raises the emitted fluorescence.
Two-photon excitation
Two long-wavelength photons excite one fluorophore. Absorption is non-linear in photon density, so only a single voxel is excited at a time — that is what gives optical sectioning in intact tissue.
Types of errors in imagery
Section/alignment artifacts (folds, cracks, missing sections) that propagate into segmentation.
Ultrastructure
Fine sub-cellular EM features: organelles, mitochondria, synapses, myelin.
Unit
Two different recording modalities use this word for their basic recorded element, and they are not the same thing.
Unit quality metrics
Per-unit numbers describing how badly spike sorting may have gone wrong for that unit — contamination from other neurons, spikes missed, or the unit drifting away. None is perfect; which thresholds apply depends on the analysis.
Units table
One row per sorted unit: spike times, mean waveform, quality metrics, and the peak channel that gives it a location. The primary table of any ephys dataset.
Unproofread
An arbor that has not been comprehensively corrected. It is truncated by split errors and may carry merged fragments of other cells, so its apparent partners are unreliable.
V1DD (V1 Deep-Dive)
Functional (2p/3p calcium) + EM dataset of V1 across all layers in 4 mice (~50k neurons/mouse).
V1DD functional index
V1DD's Golden-Mouse column/volume/plane/roi scheme, distinct from MICrONS session/scan/unit.
valid_roi
The ophys equivalent of a unit quality flag: whether cell classification judged a segmented ROI to be a real cell. Only valid ROIs are released.
Vertex / Vertices
Points in 3D (N×3, nanometers) that, connected, build meshes and skeletons.
VIP cell
An inhibitory interneuron expressing Vasoactive Intestinal Protein. VIP cells tend to target somatostatin cells rather than excitatory neurons; this role as a “disinhibitory specialist” is thought to matter for context-dependent modulation of cortical activity.
VISp / VISal / VISrl
The visual cortical areas (V1 / AL / RL / LM) the volume spans and assigns.
Volume
A cubic-mm 3D EM image dataset spanning a cortical region.
VORTEX
NIH program (Virtual Observatory of the Cortex) funding continued proofreading; source of the vortex_* tables.
Voxel
The smallest 3D image unit; anisotropic 4×4×40 nm (MICrONS) / 9×9×45 nm (V1DD).
Watertight
EM meshes are NOT watertight, so Trimesh .volume/.center_mass are invalid.
Waveform
The voltage over time measured at an electrode when a neuron fires an action potential. The per-unit mean waveform is what the shape metrics are computed from.
ΔF/F (dF/F)
Change in fluorescence normalised by a baseline. The baseline is the median fluorescence in a 180 s window centred on each time point, so ΔF/F is a relative, unitless signal.
Nothing matches that search.
Generated from the Allen Glossary
(revision 2026-08), which is the source of truth for these definitions —
corrections and new terms belong there, not on this page.
Further reading: MICrONS Explorer · CAVEclient documentation · NWB · AIND open data on S3
Term index#
The same 249 terms as a plain list, A to Z. This is what the databook’s own
search box and any {term} cross-reference elsewhere in the book resolve against,
so it is folded away rather than left out.
Every term, A to Z
- 3D reconstruction#
Turning EM imagery into 3D neuron objects (dense segmentation → meshes). Go to the card.
- Action potential#
- Spike#
A characteristic signal in excitable cell membranes: a potential-difference waveform that propagates along the membrane. In neurons it indicates activation. The trace is a Hodgkin-Huxley simulation: a brief current pulse opens sodium channels, which depolarise the membrane and then inactivate, while potassium conductance rises more slowly and repolarises it past rest. Go to the card.
- AIND metadata schema#
Six JSON classes describing a newer data asset: data description, subject, procedures, rig or instrument, session or acquisition, and processing. Where you look up which virus was injected, or what a capsule actually ran. Go to the card.
- AllenSDK#
The Python package for the Brain Observatory physiology datasets, wrapping downloads and metadata behind a cache object. Being retired in favour of reading NWB files directly, so new work should not start here. Go to the card.
- amplitude_cutoff#
Estimated fraction of the unit’s spikes that fell below the detection threshold and were never recorded — a false-negative rate. Default threshold 0.1. Go to the card.
- Annotation#
Labeled data (points/tables) bound to locations or cells in the volume. Go to the card.
- Baiting / coupled vs uncoupled#
Baiting: a reward an unchosen side would have given is held and delivered on the next choice of that side. Coupled or uncoupled describes whether the two sides’ probabilities change together or independently. Go to the card.
- Basket cell (BC)#
- Basket cell#
Inhibitory neuron whose synaptic output targets the cell body and proximal dendrites of excitatory neurons. Many basket cells express parvalbumin (PV), but not all — some express cholecystokinin (CCK). PV basket cells are typically fast spiking, and are thought to be important for gain control and for the temporal precision of network activity. Go to the card.
- BCI task / conditioned neuron#
A lickport moves toward the mouse at a speed set by the fluorescence of one chosen neuron. Reaching the near position within 10 s earns water. Mice usually learn to drive that neuron within about 30 trials. Go to the card.
- Behavior session#
One behavioural recording, whether it happened under the microscope or in the training facility. Its session_type names the training stage, which is how the full training history is reconstructed. Go to the card.
- Bipolar cell (BPC)#
- Bipolar cell#
A subset of VIP cell with a bipolar dendritic arbor — two primary dendrites leaving opposite poles of the soma. Distinct from the retinal cell of the same name. Go to the card.
- Blank sweep#
A trial in which the stimulus is replaced by mean-luminance grey, interleaved among real trials so each stimulus has its own baseline. Go to the card.
- Bound Spatial Point#
Binds an annotation to the cell at a location via the triad pt_position → pt_supervoxel_id → pt_root_id. Go to the card.
- Branch / End / Root point#
Named skeleton vertex types; the root is conventionally placed at the soma. Go to the card.
- Catch trial / sham change#
A change time is drawn but the image does not change. This conservative definition counts only presentations drawn from the change-time distribution; aborted trials are arguably catches too. Go to the card.
- CAVE#
Connectome Annotation Versioning Engine — the suite managing large dynamic connectomics data. Go to the card.
- CAVEclient#
The main Python client for programmatic access to CAVE services. Servers: MICrONS global.daf-apis.com, V1DD global.em.brain.allentech.org. Go to the card.
- cc_abs / cc_max / cc_norm#
Digital-twin model-performance columns. Go to the card.
- Cell type#
Classification of a cell (e.g. 23P, BC) via several tables/methods, keyed on nucleus id. Go to the card.
- cell_id / soma_id#
The 6-digit nucleus id (from nucleus_detection_v0), static across versions; tracks a cell over time. Go to the card.
- cell_specimen_id vs cell_roi_id#
cell_roi_id identifies a segmented ROI within one experiment, before matching. cell_specimen_id identifies the cell after matching across sessions, and is therefore shared across a container. Joining on the wrong one silently loses the across-day link. Go to the card.
- Change detection task#
A go/no-go task: images are presented in a continuous stream and the mouse earns water by licking when the image identity changes. The 500 ms grey gap between images adds a working-memory component. Go to the card.
- Channelrhodopsin (ChR2)#
A light-gated ion channel used in optogenetics to control neuronal activity with light. Go to the card.
- Channels table#
One row per recording site, at general/extracellular_ephys/electrodes, with its position on the shank and in the CCF. A unit points into this table through its electrodes column; that is how a spike acquires a place in the brain. Go to the card.
- ChRmine#
A red-shifted opsin, excited near 1080 nm. Because GCaMP is excited near 920 nm the two can be driven independently, which is what makes simultaneous imaging and single-cell photostimulation possible. Go to the card.
- classification_system column#
The E / I / non-neuron grouping column in cell-type tables. Go to the card.
- Clean#
Arbor proofread to remove all merge errors (synapses correct, but may be incomplete). Go to the card.
- cloud-volume / ImageryClient#
Serverless clients to read Precomputed imagery/segmentation and download aligned cutouts. Go to the card.
- Column (MICrONS)#
- Minnie column#
A 100 µm-square region spanning all cortical layers, densely proofread for a cell-type census. Go to the card.
- Column (V1DD field)#
A column field naming one of 5 stacked scan sub-volumes tiling the V1DD block — a different concept from the MICrONS column. Go to the card.
- Common Coordinate Framework (CCF)#
- CCF#
A standard 3D reference space for the mouse brain that lets data from different modalities be placed in the same coordinates. Go to the card.
- Compartment labels#
SWC integer codes: 0 undefined, 1 soma, 2 axon, 3 basal dendrite, 4 apical dendrite. Go to the card.
- Connectivity Viewer#
Dash app showing a cell’s synaptic inputs/outputs grouped and colored by cell type. Go to the card.
- Connectome#
A wiring map of neurons and the synaptic connections between them. Go to the card.
- Container#
There is no consistent use of this term. Go to the card.
- Context block#
A ten-minute stretch in which only one modality is rewarded, signalled by instruction trials at its start. Blocks alternate for six blocks in a session. Go to the card.
- Coordinate frames#
Three systems: voxel (annotations), nanometer (mesh/skeleton vertices), transformed (pia-flattened microns). Go to the card.
- Coregistration#
Aligning functionally-imaged cells to the same cells in the EM volume (manual + automatic). Go to the card.
- Cre line#
Cre recombinase catalyses recombination between loxP sites. Paired with a loxP reporter line it drives the reporter’s expression, and because Cre is expressed within a specific gene the expression is restricted to a subset of cells. Go to the card.
- ctr_pt_position#
The synapse-junction center point (not root-id-bound). Go to the card.
- Current source density (CSD)#
The second spatial derivative of the LFP along the probe, which localises current sinks and sources and so the laminar position of synaptic input. Go to the card.
- d_prime (unit)#
Separability of this unit’s waveforms from its neighbours’, by linear discriminant analysis. Higher is better. Not the behavioural d-prime. Go to the card.
- d-prime (behavioural)#
Signal-detection sensitivity for the task: how far the hit rate exceeds the false-alarm rate. Not the unit quality metric of the same name. Go to the card.
- Dash web apps#
Plotly-Dash apps (Table Viewer, Connectivity Viewer) for fast querying + Neuroglancer-link generation. Go to the card.
- Datastack#
A named bundle of imagery + segmentation + annotation DB (minnie65_public, v1dd_public). Go to the card.
- decoder_label#
The pipeline’s automated call on what a unit is — sua for a single unit, and so on — with decoder_probability as its confidence. Go to the card.
- Default quality filtering#
Visual Coding applies isi_violations, amplitude_cutoff and presence_ratio filters by default; Visual Behavior Neuropixels returns every unit unfiltered. Same SDK, opposite defaults — check which you are holding. Go to the card.
- default_qc#
A single pass/fail flag summarising the pipeline’s quality criteria for a unit, in the AIND-packaged datasets. Go to the card.
- Depth / pia→WM axis#
y increases with cortical depth, so depth plots need ax.invert_yaxis(). Go to the card.
- Digital twin#
A DNN trained to predict a cell’s response to arbitrary stimuli (source of derived functional properties). Go to the card.
- Direct vs indirect activation#
The central pitfall of optotagging: a neuron may respond to the laser because it expresses the opsin, or because a neuron that does synapses onto it. Direct responses are short-latency (<10 ms), reliable across pulses, and tightly distributed in time. Go to the card.
- Distance#
Four geometric senses and two statistical ones are in routine use, and they give different answers for the same pair of points. Go to the card.
- Drift metrics#
max_drift and cumulative_drift record how far, in µm, a unit’s spikes moved along the probe during the session. Newer pipelines add activity_drift and drift_ptp. Go to the card.
- Drifting gratings#
A full-field sinusoidal grating moving orthogonal to its own orientation. Parameters: orientation and direction (degrees), temporal frequency (Hz), spatial frequency (cycles/deg), contrast. Typically 2 s on, 1 s grey. Go to the card.
- Driver line#
A transgenic line engineered to label a specific cell population by expressing a gene under that population’s promoter. The driver line determines which cells are targeted; the reporter line determines what is expressed in them. Go to the card.
- DSI#
Direction selectivity index (0–1). Go to the card.
- Dynamic foraging task#
Two choices, binary reward, and reward probabilities that change during the session. A go cue opens a short window in which the mouse licks left or right; the mouse must learn from recent outcomes to track the better side. Go to the card.
- Dynamic Routing task#
A context-dependent go/no-go task alternating visual and auditory blocks. The same stimulus is a target or not depending on the current block, so stimulus and meaning can be separated. Go to the card.
- Edges#
Pairs of connected vertices (mesh.edges, skeleton edges). Go to the card.
- Electron microscopy (EM)#
Imaging that reaches nanometer resolution to reveal tissue ultrastructure. Go to the card.
- Encoding vs decoding#
Encoding asks whether an event changes neural activity; decoding asks whether the event can be read back out of the activity. Same data, opposite direction. Go to the card.
- Enhancer AAV#
A virus carrying a cell-type-specific enhancer, used to restrict expression without breeding a transgenic line. Go to the card.
- Environment secrets#
How the CAVE auth token is supplied when code runs on a shared or hosted machine: exported as environment variables named API_SECRET_
instead of being written to a credentials file in the home directory. Go to the card. - Ephys#
Shorthand for electrophysiology. Go to the card.
- Ephys selection bias#
Spike sorting needs enough spikes to form a cluster, so sparsely active neurons are missed and large-spike, high-rate neurons — and layer 5 — are over-represented. Ophys sees many of the cells ephys does not. Go to the card.
- Epoch#
A labelled stretch of time — but of what, and on whose clock, differs everywhere it appears. Go to the card.
- Error profiles#
The characteristic ways automated segmentation fails, and how they differ by compartment: thin axons are dominated by split errors, thicker dendrites and somata by merges. This asymmetry is why proofreading status is tracked separately for axon and dendrite. Go to the card.
- Event detection#
Deconvolving ΔF/F into discrete events, here with the L0 method. At population imaging resolutions 1- and 2-spike events are detected unreliably, particularly with GCaMP6f. Go to the card.
- Evoked vs spontaneous#
Activity driven by a stimulus versus activity during the grey-screen epochs. The comparison that decides whether a response is a response at all. Go to the card.
- Excitatory V1 cell types#
Pyramidal subclasses by layer/projection: 23P, 4P, 5P-IT/ET/NP, 6P-IT/CT (+ mtype clusters L2a…L6wm). Go to the card.
- Experience level#
Whether the image set in a session is the one the mouse trained on (Familiar) or a different one (Novel). The axis the Visual Behavior datasets were built to test. Go to the card.
- Experiment#
There is no consistent use of this term. Establish which one is meant before joining anything. Go to the card.
- Extended#
Arbor proofread to remove all merge AND split errors (correct and as-complete-as-possible). Go to the card.
- Extracellular electrophysiology#
Recording voltage from outside the cell membrane, which gives better access to intact brains than intracellular recording. Its two readouts are spikes and the local field potential. Go to the card.
- Eye tracking / pupil#
Ellipse fits to eye, pupil and corneal reflection per video frame, giving area, centre and rotation, plus a likely_blink flag. Recorded during physiology sessions but not during training. Go to the card.
- Faces#
Triangles of connected vertex indices that tile a mesh surface (mesh.faces). Go to the card.
- Fast spiking neuron (FSN)#
Narrow, fast action potentials; with enough injected current, high spike rates without frequency adaptation. In unlabelled extracellular recordings, narrow-waveform units are called fast spiking and putatively identified as PV+ cells. Go to the card.
- FIBSEM#
Focused-ion-beam SEM; block-face EM that mills & images, giving near-isotropic voxels. Go to the card.
- Field of view#
The imaged extent of one plane, in pixels and in µm. Recorded per experiment as field_of_view_width/height. Go to the card.
- firing_rate#
Mean spike rate over the whole session. Low values may mean a sparsely active neuron or a badly detected one. Go to the card.
- Fluorophore#
A molecule that absorbs light and re-emits it at a longer wavelength. Fluorophores fluoresce only while exposed to a light source. Go to the card.
- Functional connectome#
A dataset linking synapse-resolution EM connectivity to recorded neural function in the same neurons. Go to the card.
- GABA#
The main inhibitory neurotransmitter in the mammalian brain. In cortex most GABAergic neurons are local interneurons. Go to the card.
- Gabor patches#
Spatially restricted gratings. The receptive-field mapping stimulus in Visual Coding Neuropixels: 20° diameter, three orientations on a 9 × 9 grid of screen positions, identical in every session. Go to the card.
- GCaMP#
A family of GECI fusing calmodulin’s calcium-binding domain to green fluorescent protein. GCaMP6f and 6s are the fast and slow variants, differing in sensitivity and especially in decay kinetics. Go to the card.
- Genetically-encoded calcium indicator (GECI)#
- GECI#
A protein expressed by a cell that changes its fluorescence on binding Ca²⁺, used to visualise neural activity with fluorescence microscopy. Go to the card.
- Golden Mouse (409828)#
The single V1DD mouse with functional coregistration. Go to the card.
- gOSI / gDSI#
Global orientation/direction selectivity indices (vector-sum variant). Go to the card.
- Graphene (graphene://)#
URL protocol for dynamic, CAVE-backed (editable) segmentation/meshes, vs static precomputed://. Go to the card.
- Graphene vs Precomputed#
graphene:// = dynamic/editable; precomputed:// = static. Go to the card.
- Grids / Chunk#
The volume is partitioned into a 3D grid of chunks for the chunked-graph. Go to the card.
- Head fixation / head bar#
A surgically implanted bar clamps the mouse’s head in a repeatable position — better than 10 µm across clamp cycles, which is what makes it possible to return to the same cells on a later day. Go to the card.
- Higher visual area (HVA)#
- HVA#
A cortical visual area receiving input from primary visual cortex, and so higher in the visual hierarchy. In the mouse: VISl, VISal, VISpm, VISam, VISrl among others. Go to the card.
- Hit / miss / false alarm / correct reject#
Lick within the 750 ms window after a change = hit; no lick after a change = miss; lick after a sham change = false alarm; withholding on a sham change = correct reject. Licking before the scheduled change aborts the trial. Go to the card.
- Image set#
Which eight natural images a session used (G or H, A or B). Two images are shared between sets, so novelty is a property of the other six. Go to the card.
- Imagery#
The 3D grayscale (0–255) array depicting EM ultrastructure. Go to the card.
- Imaging depth#
Depth in µm below the cortical surface at which a plane was collected. Roughly: <250 layer 2/3, 250–350 layer 4, 350–500 layer 5, >500 layer 6 — but layer-specific Cre lines are the reliable way to get layer specificity. Go to the card.
- Imaging plane#
One two-photon focal plane. A single-plane microscope images one per session; the Multiscope/Mesoscope images up to eight. The plane, not the session, is what an ophys experiment is defined on. Go to the card.
- Indicator sparsification#
Calcium indicators respond non-linearly to firing rate: bursts are boosted, isolated spikes washed out. Tuning measured with ophys therefore looks sharper and sparser than the same tuning measured with ephys. Go to the card.
- Inhibitory V1 cell types#
Interneuron subclasses: BC, BPC, MC, NGC (manual) and PTC/DTC/STC/ITC (targeting-based mtypes). Go to the card.
- Interneuron#
A neuron with short axons that synapses only with nearby neurons. In cortex the term is often used to mean an inhibitory neuron. Go to the card.
- Intrinsic signal imaging (ISI)#
- ISI#
Measuring blood-flow changes from the reflectance of red light on the brain surface. Commonly used to map retinotopy across the cortical surface and so to target later recordings. Go to the card.
- isi_violations#
Rate of inter-spike intervals shorter than the refractory period. A real neuron cannot fire that fast, so violations mean spikes from more than one cell were merged. Default threshold 0.5. Go to the card.
- isolation_distance#
Distance in Mahalanobis space to the nearest other cluster of waveforms. Higher is better separated. Go to the card.
- IT / ET / NP / CT / SP#
Projection categories: intratelencephalic, extratelencephalic, near-projecting, corticothalamic, subplate. Go to the card.
- Kilosort#
The template-matching sorter used for all Allen Neuropixels data. It merges automatically, so no manual curation step is needed for recordings with little drift. Go to the card.
- l_ratio#
Contamination measure related to isolation distance: the probability that nearby spikes belong to this cluster. Lower is better. Go to the card.
- Layer (cortical)#
L1–L6 along the pia→WM axis; drives cell-type naming. NOT the Neuroglancer layer. Go to the card.
- Level of detail (LOD)#
Static meshes are smaller, multi-LOD, precomputed://; dynamic meshes are detailed, single-LOD, graphene://. Go to the card.
- Local field potential (LFP)#
- LFP#
- Local field potential#
Transient electrical potential generated in nervous tissue by the summed activity of the cells in it, typically measured below 250 Hz. Informative about oscillations and network synchrony. Go to the card.
- Locally sparse noise#
Black and white spots flashed on a grey screen, arranged so no two spots fall within 5 pixels of each other. The exclusion zone is what makes the average around any pixel structureless, so a receptive field can be recovered. Go to the card.
- Manifest#
The file a cache uses to know what data exists and where it was put. Instantiating a cache without naming one creates it in the working directory. There is no manifest when you read NWB directly; the file is the manifest. Go to the card.
- Martinotti cell (MC)#
- Martinotti cell#
A subtype of SST cell that targets the apical dendrites of pyramidal cells in layer 1. Martinotti cells are found in layer 2/3 and layer 5. Go to the card.
- Materialization & Versioning#
Timestamped snapshots of the annotation DB; each version = a fixed timestamp (MICrONS v1507, V1DD v1196). Go to the card.
- Maximum / average projection#
The imaging movie collapsed over time into one image — the standard way to see every cell in a plane at once. Go to the card.
- Merge errors#
Two neurons’ processes incorrectly joined; they add false connections. Go to the card.
- Meshes#
Vertices + triangular faces defining a neuron’s 3D outer surface. Go to the card.
- MeshParty / Meshwork#
Python package + object bundling the L2 mesh, skeleton, and anno annotations, kept in sync. Go to the card.
- Meshpoints#
Informal usage for mesh vertices. Not a formal term — say vertices, since “point” elsewhere means an annotation position. Go to the card.
- MICrONS#
Cubic-millimeter functional-connectomics EM dataset of mouse visual cortex (VISp/VISal/VISrl). Go to the card.
- Minnie#
Internal name for the MICrONS dataset/mouse (minnie65; datastack minnie65_public). Go to the card.
- Motion correction#
Registering every frame of the imaging movie to a reference before segmentation, so an ROI mask refers to the same cell throughout. Go to the card.
- mtypes#
Morphology/connectivity-derived cell-type clusters (L2a…L6wm; PTC/DTC/STC/ITC). Go to the card.
- Natural movies#
Black and white film clips with natural spatial and temporal statistics — usually the opening shot of Touch of Evil, chosen because it is continuous, with no cuts and varied motion. Go to the card.
- Natural scenes#
Black and white photographs with natural spatial statistics, flashed for 0.25 s with no gap. Visual Coding uses 118 images drawn from the Berkeley, van Hateren and McGill image sets. Go to the card.
- Neuroglancer#
WebGL browser viewer for very large volumetric connectomics data (imagery, segmentation, meshes, annotations). Go to the card.
- Neuroglancer forks#
Neuroglancer is maintained as several diverging branches. Spelunker is the one CAVE datastacks link to; the Seung-lab and FlyWire branches are the other widely used ones. States are broadly compatible but not identical. Go to the card.
- Neuroglancer Layer (img/seg/ann)#
The data layers in a Neuroglancer state. NOT the cortical layer. Go to the card.
- Neuroglancer State#
JSON object storing all layers/view/annotations, identified by a state id. Go to the card.
- Neurogliaform cell (NGC)#
An interneuron that makes a diffuse axonal arbor and is thought to release GABA through both synaptic release and volume transmission, non-selectively inhibiting nearby neurons. Go to the card.
- Neuronal process#
An axon or dendrite branch of a neuron (a process that splits at branch points). Go to the card.
- Neuropil correction#
An annulus around the ROI, excluding nearby cells, gives a local neuropil signal. It is subtracted from the raw trace after weighting by a per-cell r value. Go to the card.
- Neuropixels#
A family of silicon probes for high-channel-count single-unit extracellular recording, miniaturised with integrated-circuit design so that hundreds of units can be recorded from one probe with minimal brain damage. Go to the card.
- nglui (statebuilder/parser)#
Python package to generate and parse Neuroglancer states from dataframes. Go to the card.
- nn_hit_rate / nn_miss_rate#
Nearest-neighbour estimates of contamination and of missing spikes respectively. Go to the card.
- Nodes#
Vertices in the skeleton / L2 graph. Go to the card.
- NP 1.0 / 2.0 / Ultra / Opto#
1.0: 960 sites, ~20 µm pitch, ~3.8 mm span. 2.0: 1280 sites per shank, ~15 µm pitch. Ultra: 6 µm pitch, fine detail over a shorter span. Opto: 1.0 plus 28 on-shank light emission sites. All read out 384 channels at a time. Go to the card.
- NWB (Neurodata Without Borders)#
The standard file format for physiology and behaviour data. Visual Coding and Visual Behavior use an HDF5 backend; the newer datasets — V1DD, BCI, Dynamic Foraging, NP Ultra — use a Zarr backend optimised for cloud access. Go to the card.
- NWB layout#
Every NWB file has the same top-level groups: general (subject, devices, electrodes or imaging planes), acquisition (signals as acquired), stimulus (what was presented), intervals (epochs, trials, blocks), processing (anything derived), units (sorted units, ephys only) and analysis (non-standard extras). What differs between datasets is what fills them — and where a dataset puts a thing is not always where you would guess, so print the tree first. Go to the card.
- Omission#
5% of non-change presentations are dropped, interrupting the expected stimulus cadence so that expectation signals can be measured. Omissions occur during recording but not during training, and never at or just before a change. Go to the card.
- Ophys#
Shorthand for optical physiology, often in reference to two-photon calcium imaging, but can also include other methods such as fiber photometry. Go to the card.
- Ophys container#
The same imaging plane followed across days. Containers hold different numbers of sessions depending on which passed QC and how many retakes happened. Go to the card.
- Ophys experiment#
One imaging plane in one session — the narrowest unit in the hierarchy, with its own imaging_depth and targeted_structure. Quality control passes or fails each plane separately. Go to the card.
- Ophys session#
One continuous recording under the two-photon microscope. It contains one imaging plane on a single-plane scope and up to eight on the Multiscope. Go to the card.
- Opsin#
A light-gated ion channel. Illumination changes its conformation, letting ions cross the membrane and either forcing the cell to spike (excitatory opsin) or suppressing spiking (inhibitory). Go to the card.
- Optogenetics#
Controlling neural activity by expressing light-activated ion channels in a specific subpopulation — a reporter line for the opsin, a driver line for the population — giving temporally precise control of spiking. Go to the card.
- Optotagging#
Using optogenetics to identify which recorded units belong to a genetically defined population, by their response to laser pulses. Trains of 10 ms pulses at 20 Hz are a common stimulus. Go to the card.
- Oracle score#
Visual-response reliability — signal correlation across repeated “oracle” movies. Go to the card.
- OSI#
Orientation selectivity index (0–1). Go to the card.
- Parvalbumin-positive (PV+) neuron#
- Parvalbumin-positive interneuron#
Fast-spiking GABAergic interneurons with strong inhibitory effects on their neighbours; action potentials can be under 400 µs. Parvalbumin is a calcium buffer, so calcium imaging of these cells should be read cautiously. Go to the card.
- Passive replay block#
The same stimuli replayed with the lick spout retracted and no reward, so task-dependent modulation can be separated from stimulus drive. Go to the card.
- Peak channel#
The channel on which a unit’s mean waveform is largest. A unit carries no position of its own — joining peak_channel_id to the channels table is how it acquires a CCF location, a brain-region label and a depth. Go to the card.
- Physiology#
The activity side of a functional-connectomics dataset: the calcium-imaging responses recorded from the same neurons that were later reconstructed in EM. Go to the card.
- Position#
The 3D coordinate of a bound spatial point (pt_position, stored in voxels by default). Go to the card.
- Precomputed format#
Storage representation for arbitrarily large images/meshes/skeletons. Go to the card.
- pref_dir#
Preferred direction in degrees (0–360; 0 = vertical bar moving right, CCW+). Go to the card.
- pref_ori#
Preferred orientation in degrees (0–180). Go to the card.
- presence_ratio#
Fraction of the session in which the unit had spikes. A low value usually means the unit drifted away from the probe. Default threshold 0.9. Go to the card.
- Probe / shank / channel / site#
The recording hierarchy: a probe carries one or more shanks, a shank is patterned with recording sites, and the subset wired out for recording at any moment are the channels. Go to the card.
- Project cache#
The AllenSDK entry point for the Brain Observatory datasets: it downloads what you ask for, keeps it in a known directory, and hands back manifest tables and session objects. Newer datasets have no cache — you open the NWB file yourself. Go to the card.
- Proofreading#
Manual correction of split/merge errors to make neurons biologically accurate/complete. Go to the card.
- PSTH#
Peri-stimulus time histogram: spikes binned relative to stimulus onset and averaged over trials, giving the time course of the response. Go to the card.
- PyChunkedGraph (PCG) / L2 graph#
Hierarchical representation: L0 = voxels, L1 = supervoxels, L2 = supervoxels grouped within a chunk. Go to the card.
- Pyramidal cell#
An excitatory neuron with a characteristic cell-body shape and apical dendrite. In visual cortex, by far the most common excitatory type. Go to the card.
- Q value / RPE#
Latent variables of a reinforcement-learning fit to foraging behaviour: the expected value of each choice, and the reward prediction error that updates it. Useful precisely because they can then be regressed against neural activity. Go to the card.
- query_table / synapse_query#
The two query entry points + filter_in_dict; note the 200k-row cap, desired_resolution, select_columns, split_positions. Go to the card.
- Radius#
Half the cable thickness at a skeleton vertex (µm). Go to the card.
- readout_loc_x/y#
Approximate receptive-field center in stimulus space. Go to the card.
- Receptive field#
The region of the stimulus domain in which a stimulus must lie to evoke a response. Generalises beyond space to any stimulus dimension, and so to the stimulus features that drive a cell. Go to the card.
- Reference table#
A table linked to another (usually nucleus_detection_v0) by shared annotation id, adding _ref columns. Go to the card.
- Regular spiking neuron (RS)#
Longer action potentials and spike-frequency adaptation — the rate falls over a sustained current step. The most common cortical type, usually associated with excitatory pyramidal neurons. Go to the card.
- Reporter line#
A transgenic line engineered to express a protein that monitors or manipulates activity — GFP, GCaMP, channelrhodopsin — but only once the controlling protein (Cre or FLP) is present. Go to the card.
- Residual / Separation score#
The two coregistration-quality metrics. Go to the card.
- Resolution#
Physical voxel size in nm/voxel (MICrONS 4×4×40; V1DD 9×9×45); set per query via desired_resolution. Go to the card.
- Response modulation index (RMI)#
The normalised contrast between visual and auditory target response rates, collapsing two hit rates into one number that says which context the mouse is behaving in. Go to the card.
- Retake#
A second attempt at a session_type after the first failed QC. Why prior_exposures_to_image_set and not session_type tells you whether a session was truly the first with novel images. Go to the card.
- Retinotopy#
- retinotopic map#
The mapping of visual space onto neural space: neighbouring points in the visual field are represented by neighbouring points in the brain. Measured as altitude (upper–lower) and azimuth (left–right). Go to the card.
- ROI mask#
- ROI#
The pixel mask for one segmented cell in an imaging plane. In two-photon data an ROI is the set of pixels thought to belong to a single neuron. Go to the card.
- Root_id (pt_root_id)#
Unique integer for a specific segmentation = a specific version of a cell (a.k.a. segment / object id). Go to the card.
- Running speed#
Speed on the running disc, temporally aligned to the activity traces. Same length as ΔF/F, so a stimulus epoch indexes into both. Go to the card.
- Saccade#
A rapid ballistic eye movement between fixation points. Mice are not foveal animals and their eye movements differ from those of foveal species. Go to the card.
- Scan#
The scan_idx from functional imaging; part of the ROI’s unique id. Go to the card.
- Segmentation#
A 3D array where each voxel stores the root_id of the object at that location. Go to the card.
- Segments (= root/object id)#
“Segment id” used as a synonym for root id — collides with the skeleton sense of “segment”. Go to the card.
- Segments (skeleton)#
An unbranched run of vertices between branch/end points. Go to the card.
- Serial-section EM#
Many ultrathin sections are cut from a block, imaged one by one, then re-aligned into a volume. Resolution is fine in x/y and coarse in z, so voxels are strongly anisotropic. Go to the card.
- Session#
The databook defines it as “a physiological and/or behavioral recording that happens at one time”, but four narrower senses are in use as identifiers. Go to the card.
Authenticated state-sharing mechanism. Go to the card.
- Signal vs noise correlation#
Signal correlation compares two cells’ mean responses across stimulus conditions — do they like the same things. Noise correlation compares their trial-to-trial fluctuations to the same condition — do they vary together. Go to the card.
- Single unit vs multi-unit#
Not two categories but a gradient, from complete and uncontaminated to incomplete and highly contaminated. Every analysis still has to draw a binary line somewhere; quality metrics are how you draw it deliberately. Go to the card.
- Skeletons#
Tree-like linear representation of a neuron’s branching (vertices + edges, radius, compartments). Go to the card.
- snr#
Waveform amplitude relative to background noise on the peak channel. Go to the card.
- Somatostatin (SST) cell#
- Somatostatin cell#
An inhibitory interneuron expressing somatostatin (SST, sometimes SOM). SST cells tend to target the distal dendrites of excitatory neurons, and have important roles in regulating their activity. Go to the card.
- Source#
Disambiguation: image_source/segmentation_source, the Neuroglancer layer source, and skeleton path_between(source,…). Go to the card.
- Source (presynaptic)#
The presynaptic partner of a synapse (pre_pt_root_id). Go to the card.
- Spatial frequency#
How often the sinusoidal components of a signal repeat per unit distance — for a grating, the spacing of its bars. Typically cycles per degree. Go to the card.
- Spike band / LFP band#
The two streams split off each channel: the spike band at 30 kHz with a 500 Hz high-pass, carrying action potentials from adjacent neurons; the LFP band at 2.5 kHz, carrying low-frequency fluctuations from a wider area. Go to the card.
- Spike raster#
One row per trial, one tick per spike, aligned on an event. The plot to make before any model, because it shows trial-to-trial structure that an average hides. Go to the card.
- Spike sorting#
Assigning detected spikes to individual neurons — a blind source separation problem. Detection, extraction, feature extraction, clustering, then validation against the refractory period. Go to the card.
- Split errors#
A process incorrectly appears to stop; they remove true connections. Go to the card.
- Spontaneous activity#
An epoch of mean-luminance grey with no patterned stimulus, included in most sessions as a baseline for visually evoked activity. Go to the card.
- standard_transform#
Package converting voxel/nm coordinates to pia-flattened micron coordinates (minnie_ds, v1dd_ds). Go to the card.
- State#
Four unrelated meanings, two of which appear in the same workshop. Go to the card.
- Static gratings#
A stationary full-field sinusoidal grating flashed for 0.25 s. No temporal frequency; phase becomes a parameter instead. Go to the card.
- Status flags#
Booleans status_axon/status_dendrite recording whether each arbor was proofread, plus valid_id (root id at assessment). Go to the card.
- Stimulus epoch table#
When each interleaved stimulus block began and ended. In Visual Coding 2P the bounds are given as imaging frames, so they index directly into the ΔF/F and running-speed traces. Go to the card.
- Stimulus presentations table#
One row per stimulus shown, with its parameters and its start_time and stop_time. The table every alignment starts from. In NWB it lives under stimulus/presentation, or as a TimeIntervals table under intervals — which one depends on the dataset. Go to the card.
- Stimulus template#
The literal image shown, stored alongside the stimulus table for image and movie stimuli. Often available both unwarped and warped — the warped version is what the monitor rendered. Go to the card.
- Strategy values#
dendrite_clean, dendrite_extended, axon_partially_extended, axon_fully_extended, axon_interareal (MICrONS only), none. Go to the card.
- Structure acronym#
The CCF region label attached to a channel or unit — VISp, MOs, LSr. A unit with no CCF registration gets coordinates of [-1, -1, -1]. Go to the card.
- Supervoxel (pt_supervoxel_id)#
L1 grouping of voxels within a chunk; the stable internal id an annotation binds to. Go to the card.
- Surround suppression#
A stimulus extending beyond a cell’s classical receptive field suppresses its response. Stronger in superficial layers, and one of the questions V1DD’s windowed and full-field gratings were designed to address. Go to the card.
- SWC format#
Standard skeleton file format (one of three: SWC, meshwork-h5, precomputed). Go to the card.
- Synapse size#
Synapse size in voxels; correlates with surface area / strength. Go to the card.
- synapse_target_predictions_ssa#
Per-synapse postsynaptic-compartment prediction (soma / spine / shaft). Go to the card.
- synapses_pni_2 / synapses_v1dd#
The sole synapse tables (337M / 639M rows). Go to the card.
- Table Viewer#
Dash app to query/filter one table and select rows in Neuroglancer. Go to the card.
- Tables#
CAVE annotation tables (synapses, nuclei, cell types, proofreading, coregistration). Go to the card.
- Tags / Shortcuts#
Keyboard-driven annotation labels for fast bulk labeling in Neuroglancer. Go to the card.
- Target#
Disambiguation: target_id (reference link) vs synaptic postsynaptic partner vs path target_index. Go to the card.
- Target (postsynaptic)#
The postsynaptic partner of a synapse (post_pt_root_id). Go to the card.
- TEASAR#
Algorithm that turns the L2 graph into a skeleton tree. Go to the card.
- TEM#
Transmission EM; MICrONS/V1DD are serial-section TEM-style (thin sections, anisotropic z). Go to the card.
- Temporal frequency#
How many complete periods the signal goes through per unit time. Typically Hz. Go to the card.
- Three-photon (3P) imaging#
Raises signal-to-noise for deep imaging of densely labelled tissue. Used to extend the V1DD centre column to white matter, where 2P image quality has degraded. Go to the card.
- Token / auth#
Google-account credential required before any programmatic access, saved per server. Go to the card.
- Transgenic line#
A mouse line whose genome has been altered by introducing foreign DNA. Here, typically a Cre line driving expression of a reporter line within a specific subset of cells. Go to the card.
- Trials table#
One row per trial: timing landmarks and outcome flags. Usually nwb.intervals[‘trials’], but not always — the BCI dataset keeps its trials under stimulus/presentation, because there the lickport is driven by the neuron. And a “trial” is not always behavioural: in the cell-type look-up table it is a laser pulse train. Go to the card.
- Tuning curve#
Mean response plotted against a stimulus parameter. The shape of the curve is what selectivity indices such as OSI and DSI summarise in one number. Go to the card.
- Two-photon calcium imaging#
Measuring neural activity through a fluorescent calcium indicator such as GCaMP. At rest a neuron has low calcium; when it spikes, calcium flows in, binds the indicator and raises the emitted fluorescence. Go to the card.
- Two-photon excitation#
Two long-wavelength photons excite one fluorophore. Absorption is non-linear in photon density, so only a single voxel is excited at a time — that is what gives optical sectioning in intact tissue. Go to the card.
- Types of errors in imagery#
Section/alignment artifacts (folds, cracks, missing sections) that propagate into segmentation. Go to the card.
- Ultrastructure#
Fine sub-cellular EM features: organelles, mitochondria, synapses, myelin. Go to the card.
- Unit#
Two different recording modalities use this word for their basic recorded element, and they are not the same thing. Go to the card.
- Unit quality metrics#
Per-unit numbers describing how badly spike sorting may have gone wrong for that unit — contamination from other neurons, spikes missed, or the unit drifting away. None is perfect; which thresholds apply depends on the analysis. Go to the card.
- Units table#
One row per sorted unit: spike times, mean waveform, quality metrics, and the peak channel that gives it a location. The primary table of any ephys dataset. Go to the card.
- Unproofread#
An arbor that has not been comprehensively corrected. It is truncated by split errors and may carry merged fragments of other cells, so its apparent partners are unreliable. Go to the card.
- V1DD (V1 Deep-Dive)#
Functional (2p/3p calcium) + EM dataset of V1 across all layers in 4 mice (~50k neurons/mouse). Go to the card.
- V1DD functional index#
V1DD’s Golden-Mouse column/volume/plane/roi scheme, distinct from MICrONS session/scan/unit. Go to the card.
- valid_roi#
The ophys equivalent of a unit quality flag: whether cell classification judged a segmented ROI to be a real cell. Only valid ROIs are released. Go to the card.
- Vertex / Vertices#
Points in 3D (N×3, nanometers) that, connected, build meshes and skeletons. Go to the card.
- VIP cell#
An inhibitory interneuron expressing Vasoactive Intestinal Protein. VIP cells tend to target somatostatin cells rather than excitatory neurons; this role as a “disinhibitory specialist” is thought to matter for context-dependent modulation of cortical activity. Go to the card.
- VISp / VISal / VISrl#
- Primary visual cortex#
- V1#
- VISp#
The visual cortical areas (V1 / AL / RL / LM) the volume spans and assigns. Go to the card.
- Volume#
A cubic-mm 3D EM image dataset spanning a cortical region. Go to the card.
- VORTEX#
NIH program (Virtual Observatory of the Cortex) funding continued proofreading; source of the vortex_* tables. Go to the card.
- Voxel#
The smallest 3D image unit; anisotropic 4×4×40 nm (MICrONS) / 9×9×45 nm (V1DD). Go to the card.
- Watertight#
EM meshes are NOT watertight, so Trimesh .volume/.center_mass are invalid. Go to the card.
- Waveform#
The voltage over time measured at an electrode when a neuron fires an action potential. The per-unit mean waveform is what the shape metrics are computed from. Go to the card.
- ΔF/F (dF/F)#
Change in fluorescence normalised by a baseline. The baseline is the median fluorescence in a 180 s window centred on each time point, so ΔF/F is a relative, unitless signal. Go to the card.
Note
This page is generated from 855c456 of the
Allen Glossary repository.
Do not edit it directly — edits are overwritten the next time it is regenerated.
To fix a definition or add a term, open a pull request against that repository.