Brody Lab Poisson Clicks Task Dataset Rats 2009 to 2024 Parsed

Subjects. A total of 515 male Long-Evans rats (Rattus norvegicus) between the ages of 2 months and 24 months were used for this study. Rats did not begin water restriction and training until their weight exceeded 225 g. The majority of rats in this study were pair housed. Animal use procedures were approved by the Princeton University Institutional Animal Care and Use Committee (IACUC) and carried out in accordance with National Institutes of Health standards.

Behavior. Rats were trained to perform an auditory evidence-accumulation decision task, the Poisson Clicks Task. For each session, rats were placed in a behavioral training box that itself was located within a sound attenuation chamber with active ventilation. The behavioral training box has three conical nose ports and two speakers. Each nose port is equipped with an infrared beam to detect nose poke events and a white LED that can illuminate the port. The left and right ports are equipped with a sipper tube and a solenoid valve that dispenses distilled water from a gravity fed tank. The center port has an enlarged opening in the center to facilitate the rat’s ability to hold their nose in the port.

Trials are self initiated by the rat who is cued with a LED in the center port. When they are ready to perform a trial the rat pokes his nose into the center port and holds it there. After a variable delay of silence two streams of auditory clicks play from each of two speakers positioned above the left and right nose ports. In the “location” variant of the task, each speaker plays an independent stream of randomly timed white noise clicks. In this version of the task the first click is often a stereo click played simultaneously from both speakers. When the clicks stop playing the center LED is extinguished and the rat is free to withdraw from the center port. The rat is rewarded with a small drop of water if they poke their nose into the side port associated with the speaker that played the greater number of clicks. An alternate “frequency” version of the task all clicks are stereo and consist of a 3 ms pure tone, either low 6.5 kHz or high 14.2 kHz. Here the rat must decide if more high or low tone clicks were played. In the frequency version high and low clicks are never allowed to overlap, and therefore there is no equivalent of an initial stereo click as in the location task.

The Dataset. This dataset contains the behavioral data of 515 rats trained in the Brody Lab at Princeton University between 2009 and 2024 to perform the Poisson Clicks Task. Any work that uses this dataset should cite the manuscript in which it was published. Some of the labels in this dataset use the word bup, this means the same as click. This is the parsed dataset. It does not contain all trails the rats performed including violation trials and trials with stimulus durations longer than 1 s. It also does not identify where the day breaks occur in the data. The full dataset will be made available in the future. Each file is a matlab .mat file named for the rat that provided the data and contains a single variable, ratdata:

ratdata =

parsed: [1×1 struct]

parsed_frozen: [1×1 struct]

task_type: 'location'

The task type identifies which version of the Poisson Clicks Task the rat trained on, Location or Frequency. The parsed field contains all trail data from unique, i.e. non-frozen trials.

ratdata.parsed =

hh: [1 × n double]

bt: {1 × n cell}

nL: [1 × n double]

nR: [1 × n double]

sd: [1 × n double]

gr: [1 × n double]

ga: [1 × n double]

rg: [1 × n double]

Where n is the number of non-violation non-frozen trials the rat performed.

hh - is a vector identifying which trials the rat got correct: 0 if they got it wrong, 1 if they got it correct

bt - is a cell which contains information about the stimulus for each trial

nL - is a vector identifying how many left evidence clicks each trial contains

nR - is a vector identifying how many right evidence clicks each trial contains

sd - is a vector of the stimulus duration in seconds for each trial

gr - is a vector identifying which response the rat made: 0 if they responded left, 1 if they responded right

ga - is a vector of the gamma value used to generate the stimulus on each trial

rg - is a vector identifying how the reward was assigned on each trial: 0 if it was offered with the side that played more clicks, 1 if it was offered on the side with the higher generative Poisson click rate

For the rats in this dataset where reward followed the trial’s gamma value, 98.5% of the trials the reward was still offered on the side that played the greater number of clicks.

At a minimum bt is a structure that has two fields:

ratdata.parsed.bt{1} =

left: [1 x nL double]

right: [1 x nR double]

left - a vector of the left evidence click times in seconds relative to the stimulus onset

right - a vector of the right evidence click times in seconds relative to the stimulus onset

Where nL is the number of left evidence clicks and nR is the number of right evidence clicks. Additional fields can include:

real_T - the duration of the stimulus in seconds

bup_width - the duration of each click in milliseconds

base_freq - the lowest frequency used to make in each click sound in Hertz

bup_ramp - defines the taper that smoothes the edges of each click

first_bup_stereo - 1 if the first click is a stereo click, 0 if not

avoid_collisions - 1 if click trains are selected where clicks do not overlap, 0 if only Poisson statistics are used to select click times

seed - the pseudorandom list seed used to generate the stimulus

is_probe_trial - 1 if the trial is a "probe" trial. These typically are defined to have a specific stimulus duration, for example 1 second, such that this duration is oversampled. Probe trials exist to facilitate other experiments such as optogenetics, however no trials with optogenetic stimuli are included in this dataset

is_frozen - 1 if this is a frozen noise trial, 0 if it is not

tones - the specific frequencies used to construct the click in Hertz

In the location version of the task typical settings are:

bup_width: 3

base_freq: 2000

bup_ramp: 2

first_bup_stereo: 1

avoid_collisions: 0

tones: [2000 4000 8000 16000 32000]

Whereas in the frequency version of the Poisson Clicks Task typical settings that differ are:

base_freq: [6500 14200]

first_bup_stereo: 0

tones: [6500 14200]

The base frequency and tone fields identify the frequency of the tone used to identify left and right evidence clicks, here 6.5 kHz is the tone frequency for a left evidence click and 14.2 kHz is the tone frequency for a right evidence click. In the frequency task the first click is typically not a stereo click.

The parsed_frozen field contains similar information as the parsed field but for the frozen noise trials:

ratdata.parsed_frozen =

hh: [1 × n double]

bt: {1 × n cell}

nL: [1 × n double]

nR: [1 × n double]

sd: [1 × n double]

gr: [1 × n double]

ga: [1 × n double]

rg: [1 × n double]

seed: [1 × n double]

unique_seed: [1 × n_uf double]

gr_seed: [1 × n_uf double]

Where n is the number of frozen noise trials the rat performed and n_uf is the number of unique frozen noise trials performed.

seed - is a vector of the pseudorandom list seeds used to generate the stimulus for each trial

unique_seed - is a vector of the unique seeds used to generate the frozen noise trials

gr_seed - is a vector of the fraction of right responses the rat makes given each unique seed

The field names in this dataset are abbreviations:

hh - hit history, hit being a trial the subject got correct

bt - bup times, bup mean the same as click

nL - number of Left

nR - number of Right

sd - stimulus duration

gr - go right, i.e. did the subject respond right

ga - gamma

rg - reward gamma