BrainsOnBoard/paper_RPEs_in_drosophila_mb

Code and data for the manuscript "Learning with reinforcement prediction errors in a model of the Drosophila mushroom body".

This folder contains all code needed to produce the simulation results and the data analyses for the paper:
"Learning with reinforcement prediction errors in a model of the Drosophila mushroom body"
Authors: Bennett JEM, Philippides A, Nowotny T

Code Contents:
 - mb_vs.m: code to run multiple alternative forced choice simulations using the VS/VSlambda model
 - mb_mv_a.m: code to run multiple alternative forced choice simulations using the MV model with overlapping cue representations.
 - mb_mv_b.m: code to run multiple alternative forced choice simulations using the MV model with overlapping cue representations, in order 
   to examine how each single Kenyon cell contributes to the reward prediction for a given cue.
 - mb_mv_c.m: code to run multiple alternative forced choice simulations using the MV model with non-overlapping cue representations and 
   perfect plasticity.
 - mb_mv_d.m: code to run multiple alternative forced choice simulations using the MV model with non-overlapping cue representations.
 - mb_vs_conditioning.m: code to run associative conditioning simulations using the VSlambda model
 - mb_mv_conditioning.m: code to run associative conditioning simulations using the MV model
 - mb_vs_conditioning_fels.m: code to simulate the Felsenberg et al. (2017, 2018) re-evaluation experiments using the VSlambda model
 - mb_mv_conditioning_fels.m: code to simulate the Felsenberg et al. (2017, 2018) re-evaluation experiments using the MV model
 - mb_mv_blocking.m: code to run simulate blocking experiments using the MV model.
 - mb_reward_schedules.m: code to generate a specific and parameterised reward schedule for use in the simulations
 - mb_vs_mv_analysis.m: code to run all data generation and data analysis.
 - zmy_utils.zip: a zipped package of utility scripts that are called upon for processing and plotting data. Please unzip this file and place all
   scripts in the Matlab path.
 
Data Contents:
 - associative_conditioning_data_for_VS_model.mat: simulated VS_lambda model data for Fig. 5 and Supp. Fig. 7 (generated using flag==9 in mb_vs_mv_analysis.m)
 - associative_conditioning_data_for_MV_model_eq7.mat: simulated MV model (Eq. 7) data for Supp. Fig. 7 (generated using flag==8 in mb_vs_mv_analysis.m) 
 - associative_conditioning_data_for_MV_model_eq8.mat: simulated MV model (Eq. 8) data for Fig. 5 and Supp. Fig. 7 (generated using flag==8 in mb_vs_mv_analysis.m) 
 - felsenberg_extinction_data_for_both_models.mat: simulated MV model (Eq. 8) for Supp. Fig. 8 (generated using flag==18 in mb_vs_mv_analysis.m)
 - source_data_1.xlsx: collated performance indeces and test statistics for experimental data used in Fig. 5 and Supp. Fig. 7.
 - source_data_2.xlsx: collated performance indeces and test statistics for experimental data used in Supp. Fig. 8.
 
Dependencies:
 - All code has been tested using either MATLAB R2012a on OS X 10.7, or MATLAB R2018b on Ubuntu 18.04 LTS. Certain subroutines in 
 mb_vs_mv_analysis.m make use of MATLAB’s Parallel Computing Toolbox. 
 - Depending on the version of MATLAB you are using, you may need to uncomment/comment out those lines in which the cluster of workers are 
 initialised.

Example usage:
 - First, you must place all .mat files in a directory that is included in the Matlab path, or update the path to include them.
 - mb_vs_mv_analysis.m is the main wrapper from which all results in the paper are generated.
 - Before you run mb_vs_mv_analysis, modify the string that is assigned to the variable ‘fpath’. This is the directory in which all 
 generated data, analyses, and plots will be saved. It is also the directory from which any data will be read in for analysis. 
 -  mb_vs_mv_analysis.m is divided into subroutines, labelled by the variable ‘flag’. To run, for example, subroutine 2 (produces 
 Fig. 2a-d in the paper), open MATLAB and run:
mb_vs_mv_analysis(2);
 - mb_vs_mv_analysis(8); generates MV model data for Fig. 5.
 - mb_vs_mv_analysis(9); generates VS_lambdamodel data for Fig. 5. 
 - mb_vs_mv_analysis(10); analyses the data generated under flag==8 and flag==9 (also provided in the .mat files in this repo) and produces Fig. 5.
 - A single run of, for example, the VS model (mb_vs.m) takes less than a second (using the reward schedules outlined in the paper). Some 
 of the more heavy analyses, such as subroutine 8 (generating performance indices for all 96 types of intervention in the associative 
 conditioning simulations) can take a few hours, depending on the specs of your computer.