This repository contains data and code that accompany the paper titled "Minimal cross-trial generalization in learning the representation of an odor-guided choice task".

Table of Contents

• data/takahashi2016roesch2009burton2018Valid.csv: behavioral data aggregated from three studies: Roesch et al. (2009), Takahashi et al. (2016), and Burton et al. (2018). Only valid trials (i.e., trials in which animals made a choice response, and received the reward delivery if reward was available) are included. See Data for details.
• model_code_stan/: model codes in stan
• model_fits/: model fits (posterior samples of model parameters)
• model_simulation/: model simulation results (due to space limit, we only include the aggregated results used for plotting average reward and learning curves)
• *.ipynb: analysis notebooks
• *.py: helper functions

Data

Each row in the data file corresponds to a trial. The columns correspond to (some contain redundent information):

• dataset: "roesch2009", "takahashi2016", or "burton2018"
• rat: rat index (1-22)
• session: session index for current animal
• sessionType: "leftBetterFirst" or "rightBetterFirst" (denotes which side has the better reward in the first block)
• trial: trial index in a session (note that animals did not make valid responses in some trials; those invalid trials are excluded from analyses in this paper)
• block: block index (1-4)
• blockType: "short_long", "long_short", "big_small", or "small_big" (denotes the reward type in left and right wells respectively)
• odor: "left", "right" or "free"
• choice: 1 = left, 2 = right
• rewardAmount: 0, 1 or 2
• rewardDelay: in seconds
• trialType: all are "valid"
• trialCond: reward type (short, long, big, small) + reward side (left, right)
• trialCondCode: code for trialCond (1 = big_left, 2 = big_right, 3 = small_left, 4 = small_right, 5 = short_left, 6 = short_right, 7 = long_left, 8 = long_right)
• Reaction times (RTs): all in seconds
  • odorEntryRT: time between light on and odor port entry
  • odorExitRT: time between odor delivery and odor port exit
  • wellEntryRT: time between odor port exit and reward well entry
  • wellExitRT: time between the last reward delivery and reward well exit

Models and model fitting

Models (including the hierarchical logistic regression model and reinforcement learning models) are implemented in PyStan[1], available in model_code_stan/.

Model fitting and evaluation functions are implemented in funcs_model_fit_evaluate.py. To fit the hierarchical logistic regression model, use behavior_logistic_analysis.ipynb. To fit the reinforcement learning models, use model_fitting.ipynb.

Model fitting results (posterior samples of model parameters) are saved in model_fits/.

Analyses and figures

The following notebooks reproduce the analyses and figures in the paper:

• Fig 1C: behavior_learning_curves.ipynb
• Fig 1D: behavior_logistic_analysis.ipynb
• Fig 3A, Fig S1: model_comparison.ipynb
• Fig 3B-E, Fig4C: model_parameter_analyses.ipynb
• Fig 4A,B: model_simulation.ipynb
• Fig S2: split_half_analysis.ipynb
• Fig S3: model_parameter_posterior.ipynb

[1] To execute notebooks in this repository, a working Python3 environment with PyStan installed is required.

Reference

Roesch, M. R., Singh, T., Brown, P. L., Mullins, S. E., & Schoenbaum, G. (2009). Ventral striatal neurons encode the value of the chosen action in rats deciding between differently delayed or sized rewards. Journal of Neuroscience, 29(42), 13365-13376.

Takahashi, Y. K., Langdon, A. J., Niv, Y., & Schoenbaum, G. (2016). Temporal specificity of reward prediction errors signaled by putative dopamine neurons in rat VTA depends on ventral striatum. Neuron, 91(1), 182-193.

Burton, A. C., Bissonette, G. B., Vazquez, D., Blume, E. M., Donnelly, M., Heatley, K. C., ... & Roesch, M. R. (2018). Previous cocaine self-administration disrupts reward expectancy encoding in ventral striatum. Neuropsychopharmacology, 43(12), 2350-2360.