Attentional set-shifting is a cognitive task that requires a subject to flexibly shift attention between different sensory features of a stimulus, based on continuous trial-and-error. This data set comprises in-vivo 2-photon video recordings of fluorescent calcium signals made in the prefrontal (prelimbic + infralimbic) cortex of mice performing a serial extradimensional set-shifting task. 

The task structure is as follows. On each trial, the animal is presented with a 2.5-second compound stimulus that consists of one of two possible odors (almond oil or sesame oil) and one of two possible whisker vibration patterns (a 255Hz sinusoidal vibration or a string of Poisson-distributed clicks averaging 255Hz). Whisker cues are presented simultaneously to both sets of whiskers via a low-frequency (sub-hearing-threshold) bass speaker coupled to a funnel that directs a 75db sound wave at the whiskers, vibrating them. At the end of the 2.5-second stimulus presentation window, the water-restricted animal may lick one of two water spouts (left or right) for a water reward. In a given block of trials, what determines the location of the water reward is either the identity of the whisker stimulus or the identity of the odor stimulus. The animal must learn, within a trial block, which is the correct sensory modality to attend to to locate the reward. After the conclusion of the 2.5-second stimulus presentation, the animal has a 1.5-second window in which a lick will cause the trial to end in a reward being dispensed or witheld. Trials are separated by an 8-10-second inter-trial interval, and preceded by a 0.5-second auditory trial-start cue. The task rule (whether to attend to the whisker or odor cue to locate the reward) automatically changes each time the animal reaches a criterion-level performance of 80% correct trials within a 30-trial moving window.

The data consist of recordings from 21 animals, expressing GCaMP6f in one of three cell types: pan-neuronal expression, driven by the hSyn promotor, expression in layer-V neurons projecting to the mediodorsal thalamus, and expression in layer II-V neurons projecting to the ventromedial striatum. The data consist of three data objects.

1. A 'cellType' structure, which denotes the types of cells expressing GCaMP in each recording (PFC for the pan-neuronal group, PMD for the prefrontal-mediodorsal-thalamus group, and PVS for the prefrontal-ventral-striatum group).
2. A 'features' structure, which contains labels for six variables that fully describe the task conditions for each trial:
   1. whiskerStim, which contains 0 or 1 to indicate which of the two whisker stimuli is presented on each trial. During trial blocks in which the "attend-whisker" rule is in effect, 0 indicates that the reward is located on the left, and 1 indicates the reward is located on the right.
   2. odorStim, which contains 0 or 1 for each trial, following the same logic as the whiskerStim values.
   3. choiceLR, which contains 0 or 1 for each trial to indicate whether the animal chose the left or right lick spout.
   4. reward, which indicates whether the animal correctly located (and received) a reward (1 for correct choice and reward, 0 for incorrect choice, no reward)
   5. attendWhiskerOdor, which indicates which of the two sensory modality rules is in effect for a given trial block (0 for attend-odor rule, 1 for attend-whisker rule)
   6. sessionNum. Each set of neurons was followed for multiple days to maximize the trial yield for each neuron, and the sessionNum variable indicates which of the consiecutively recorded sessions a given trial belongs to.
3. A 'spikeHistogram' structure, which contains deconvolved GCaMP6f df/f signal for each trial. The dimensions of this data for each animal are neuron x trial x timepoint. Recordings were made at a framerate of 2.89Hz, and each of the 54 timepoints in a trial correspond with a single camera frame. The trials are aligned to the last frame of the 2.5s stimulus presentation epoch. The recordings were motion-corrected using NoRMCorre, cells were located using the CNMF-E package, and cells were co-registered across sessions using CellReg.