Dataset

Allen Brain Observatory Visual Behavior

Open behavioral and two-photon / Neuropixels physiology dataset from the Allen Institute, in which transgenic mice perform a standardized visual change-detection task on natural-image streams. The atlas slice uses only the behavior side of the dataset, accessed through the Allen SDK VisualBehaviorOphysProjectCache from the public AWS Open Data S3 bucket.

Access notes

The atlas adapter reads Visual Behavior NWB files directly with h5py rather than depending on allensdk, because allensdk pins scipy<1.11 which conflicts with the rest of the project's scipy requirements. NWB files are downloaded from the public visual-behavior-ophys-data AWS bucket via HTTPS into an ignored local cache; the slice pins one behavior session and records the NWB URL, file SHA-256, byte size, and resolved behavior_session_id in provenance.

Data formats

  • NWB:N (Neurodata Without Borders) HDF5 files
  • h5py-readable trials table at /intervals/trials

Expected trial-table mapping

Canonical fields the harmonizer expects this dataset to populate.

Canonical field Source mapping
choice Mapped from licking behavior; lick within response window is go, withhold is withhold, aborted (lick before change) is no-response.
correct True for hits and correct rejections, False for misses and false alarms.
evidence_strength None; the evidence is a categorical image transition.
feedback reward when hit or auto-rewarded, otherwise none.
response_time response_latency seconds after the change time.
reward reward_volume from the trial row when present.
session_id BehaviorSession behavior_session_id cast to string.
stimulus_side Always none; stimuli are presented centrally.
stimulus_value None; image change is categorical.
subject_id BehaviorSession metadata mouse_id.
task_variables Preserves initial_image_name, change_image_name, change_time_in_trial, is_change, is_catch, aborted, auto_rewarded, image_set, and trial_length.
trial_index Index of the row in BehaviorSession.trials.

Caveats

  • The dataset bundles a strict training curriculum; the slice should pin one fully-trained behavior session and record the training stage explicitly rather than averaging across stages.
  • Reading the NWB file directly avoids an allensdk dependency conflict but couples the adapter to the NWB schema layout used by the Visual Behavior release; future schema revisions may require adapter updates.

Papers

Findings

Curves attached to this dataset on shared signed-evidence axes.

See all findings overlaid on shared axes.