Four rats (gh001, gh009, gh010, oh011) performed a cost-benefit decision task in which they accepted or rejected offers that varied in reward size and effort, while single units were recorded simultaneously from up to five areas of the right hemisphere:

• medial/ventral orbitofrontal cortex (MO/VO)
• anterior cingulate cortex (ACC)
• dorsomedial striatum (DMS)
• ventral pallidum (VP)
• subthalamic area (STA; labelled "stn" in the file names)

Recordings were made over 71 days between 4 December 2020 and 9 January 2022 (oh011: 4-17 Dec 2020, 14 days; gh001: 1-18 Jun 2021, 17 days; gh009: 19 Dec 2021 - 9 Jan 2022, 21 days; gh010: same period, 19 days).

70 of these days are used in the analyses (gh001 17, gh009 20, gh010 19, oh011 14). After the firing-rate criterion (>= 0.1 Hz) the analyses include 2,661 single units (MO/VO 429, ACC 505, DMS 1,184, VP 132, STA 411).

Each trial is divided into task epochs: PRE (pre-cue), CUE, NP_EX (nose-poke exit / action), CORR1 and CORR6 (corridor light gates 1 and 6) and DEL (pre-reward delay, accepted trials only).

Sampling rates

Electrophysiology 20 kHz; behavioural events (nose pokes, light gates, magazine entries) 1 kHz; overhead video and its DeepLabCut tracking 100 Hz; firing rates are binned at 25 ms.

File formats

Processed tables (inputs to the statistics and figures):

     480152_spike_times.feather - firing rates of every unit (25 ms bins, causal exponential smoothing with tau = 50 ms) per epoch and trial, accepted trials (0.9 GB)

     828817_spike_times.feather - the same for rejected trials (0.35 GB)

     382702_canonical_corr.feather - cross-validated canonical correlation analysis (CCA) between every pair of the five  areas, per session, epoch and canonical dimension, with a permutation null

     746484_canonical_corr.feather -  CCA over the decision window of accepted/rejected trials

     624290_regression.feather - decoding of reward, effort, trial number and running speed from the canonical components

     258066_regression.feather - decoder trained on accepted + rejected trials, scored separately for each

     258066_xy_table.feather - running speed per trial, epoch and bin for the accepted/rejected decoder

     781039_replacement_summary.csv summary of the replacing task variables

 Source files used to regenerate the running-speed table:

     958690_DinChannels.pbz2, 958690_FineBehaviour.pbz2, 958690_TrialBehaviour.pbz2, 958690_SpikeStats.pbz2 event channels, behavioural state sequences and trial tables of all recording stages

     809107_TrialBehaviour.pbz2 - trial-by-trial behaviour of the implanted animals

     773894_xy_tables/  - 663 DeepLabCut tracking tables (animal x/y position in pixels, 100 Hz), one per stage

Code requirements

Python 3.10; numpy 1.24.4, pandas 1.5.3, scipy 1.11.4, statsmodels 0.14.5, matplotlib 3.8.4, seaborn 0.13.2, pingouin 0.5.5, pyarrow 21.0.0, scikit-image 0.22.0, pillow 11.3.0, ray 2.49.1 (environment.yml). About 10 GB of memory for the supplemental-figures script (it loads the 0.9 GB unit table); run times are under a minute for three of the scripts and about 15 minutes for the first run of the fourth (later runs use cached results in 'output/'). Input data for the code must match the files in this dataset (names and columns as supplied). Large files (the tracking tables, DinChannels, and the two unit tables; ~5 GB) are provided separately from the code repository.