This code is associated with the paper from Grogan et al., "Effects of dopamine on reinforcement learning and consolidation in Parkinson’s disease". eLife, 2017. http://dx.doi.org/10.7554/eLife.26801
The code for analysis and model fitting used in the paper "Effects of dopamine on reinforcement learning and consolidation in Parkinson's disease".
% README.txt
% This code contains the analysis used for the paper "Effects of dopamine on reinforcement learning and consolidation in Parkinson's disease".
% We do not have ethical approval to share participants' data, so we have % also included a script to generate random data to show how the scripts run.
% You will need to copy all of the files into one folder, and run them from % there.
% Some of the figures have been disabled as they use functions developed and % licensed by someone else. I have put links to those functions in the scripts % but the analysis will run with or without those parts.
% Below is a brief description of the set up, and the list of files you % should run in which order
%% Data % There are three experiments in the paper, and they are sometimes referred to by different % names. % Experiment 1 often has no moniker, and so files without F1 or F2 in the % title are for experiment 1. % Experiment 2 is referred to as F1 in the files. % Experiment 3 is referred to as F2 in the files.
% Experiment 1 has 4 sessions, one for each of 4 medication conditions, % that all PD patients complete. Healthy Controls complete one session. % Each session has a different version of the modified Probabilistic % Selection Task (PST) run (versions A-D). The files are in the format: % P101AL1.txt (1st learning block) - version A % P101AL2.txt (2nd learning block) % P101AL3.txt (3rd learning block) % P101AM1.txt (1st memory test = 0 minutes delay) % P101AM2.txt (2nd memory test = 30 minutes delay) % P101AM3.txt (3rd memory test = 24 hours delay) % P101AN1.txt (novel pairs test = 24 hours delay) % P101BL1.txt (1st learning block) - version B % and so on for versions B, C and D. % the day1, day2 and bothDays variables set the conditions for each % session. % Controls complete one session, and do not have version numbers in their % file names.
% Experiment 2 has the same 3 learning blocks, and then novel pairs test % given immediately after learning, so there are no memory tests, and only % 2 sessions for PD patients, one ON and OFF meds, and 1 session for % controls.
% Experiment 3 had a variable number of learning blocks, dependent on their % performance compared to thresholds, so are all saved in one learning % file: % P131AL1.txt % and then a novel pairs test given immediately afterwards: % P131AN1.txt % PD patients were tested ON and OFF meds (2 sessions), and controls did 1 % session.
% The CreateFakeData.m file will generate random data that fits the format % of the data analysed in the study.
%% files to run
% Here are the orders to run the files in:
%% create fake data
%if you don't have data in the format specified, this will create fake %random data to test the scripts CreateFakeData
%% Experiment 1 analysis
DataFiles;%gets file names, sets metadata BehDataAnalysis%loads up data, processes it BehGraphs%draw figures in paper BehFilterAnalysis%run filtering analysis and draw those figures
%% Experiment 2
FDataFiles%get file names for experiments 2 and 3 F1BehDataAnalysis%load up data & analyse F1BehGraphs%draw figures
%% Experiment 3 analysis
FDataFiles%get file names and metadata for experiments 2 and 3 F2BehDataAnalysis%load up data & analyse F2BehGraphs%draw figures
%% all 3 experiments
BetweenExptsGraphs % get analysed data from each experiment, combine, figures BehWinStayAnalysis %run win-stay lose-shift analysis on each experiment, draw figure
%% Model fitting % These scripts fit Q-learning models with 1 or 2 learning rates to the % behavioural data for patients, with the same parameters for all % medication conditions, and with separate learning rates for ON and OFF % conditions during learning.
QLearnNestedAuto QLearnNestedTest