Mapping patterns of thought onto brain activity during movie-watching Data and Scripts
Description
Data and scripts for Wallace et al., 2024 for reproducibility can be found here. The study set out to map patterns of thought to brain activity during movie watching. The research experiment was exploratory in nature - so there were no explicit hypotheses. However, we set out to look at whether or not variations in thought patterns during movie watching, consisting of three 11-minute clips from Citizenfour, Little Miss Sunshine, and 500 Days of Summer, by leveraging a novel sampling technique using multi-dimensional experience sampling (mDES). We were able to accomplish this by using previously-collected fMRI data (n = 44) from the naturalistic neuroimaging database (Aliko et al., 2020) where participants were scanned while watching the films, and using a separate sample of participants (n = 120) to come into the lab to watch the same films and respond to mDES probes. Our first analysis consisted of a PCA which decomposed the experience sampling data into 4 components of thought. Based on the mDES item loadings, we labelled the thought patterns 'Episodic Knowledge', 'Intrusive Distraction', 'Verbal Detail', and 'Sensory Engagement.' We found that 'Episodic Knowledge' was most reported during 500 Days of Summer, specifically predicted better comprehension performance for this movie, and was related to activity in the visual cortex, specifically in the dorsal medial regions. 'Intrusive Distraction' was associated with reduced activity in the frontoparietal network, was highly reported during Citizenfour, and predicted poorer overall comprehension performance across the movies. 'Verbal Detail' was associated with reduced activity in the auditory cortex, most reported during Citizenfour. Lastly, "Sensory Engagement" was related to multi-modal sensory engagement in auditory and visual cortexes, predicted better overall comprehension performance, and was most highly reported during 500 Days of Summer. These results highlight the critical role that sensory systems play in the multi-modal experience of movie-watching and place important constraints on the contribution of systems in association cortex, like the default network. The raw mDES data is available, as well as the full dataset with the PCA components, gradient scores and demographic information. The relevant brain maps from the imaging analyses are also included.
Files
Steps to reproduce
The first step of the analysis after collecting the experience sampling data is to decompose the data into PCA components using ThoughtSpace. A shortcut to the ThoughtSpace GitHub webpage is included in the folder for convenience. ThoughtSpace is a Python-based toolbox for analysing experience sampling data via Principal Components Analysis (PCA) to identify common "patterns of thought." You will need to have ThoughtSpace and GitHub installed to perform the PCA analysis and to produce the corresponding word clouds found in the paper. Data files can be found in the 'Data files' folder, including raw experience sampling data, the PCA data, and a larger data frame with PCA data, gradient data, and demographic information. The brain maps generated from the FSL analysis and the seed-based analysis can be found in the 'Brain Maps' folder, including the raw maps, unthresholded maps, and thresholded map for each thought component. As well, the gradient maps and Yeo maps used in this analysis are included for convenience. Conjunction maps of the thought components of interest can also be found in this folder. Aside from ThoughtSpace and FSL, the rest of the analyses were completed using scripts in R and python and can be found in the 'Scripts' folder. The 'FSL' folder contains an example .fsf script at the individual and group level, and the corresponding EVs used in the analysis subject level and group level analyses. Within 'FSL', the 'Parameter estimates' folder contains the datasheet with the parameter estimates from our data and a corresponding R script to create the corresponding plot in the paper for visualization. Data transformation scripts (calculating averages, transform data to wide) were written in R to help format the data for the various scripts below. The scripts for each of the linear mixed models (variations in thoughts, comprehension performance, and gradients) were written in R, which provide information about the thought components with respect to their relationship to the movies, comprehension performance, and the gradients. The plots (barplots, scatterplots) that illustrate these models are also included in the script for visualization. The '3D scatterplot' folder contains a script to create the 3D figures in the paper, were written in python. The 'Wordclouds' folder contains a script to create wordclouds from the neurosynth decoding analysis to visualize the term loadings, which requires reading in a .csv file with terms in the first column and the corresponding term loadings in the second column (no headings) to create the figures. The 'Spin Test' folder consists of a Spin Test script written in python, which reads in whichever maps you want to compare against whichever gradients to produce a histogram and an output with the average gradient score for the cluster map of interest and it's p-value.