The neural representation of visually evoked emotion is high-dimensional, categorical, and distributed across transmodal brain regions

This repository contains the data and code for reproducing results in our paper: Horikawa, Cowen, Keltner, and Kamitani (2020) The neural representation of visually evoked emotion is high-dimensional, categorical, and distributed across transmodal brain regions. iScience. We investigated the neural representation of visually evoked emotions using fMRI responses to 2185 emotionally evocative short videos generated by Alan S. Cowen and Dacher Keltner (PNAS, 2017).

Python version: https://github.com/KamitaniLab/EmotionVideoNeuralRepresentationPython

Data (fMRI data and features)

• The preprocessed fMRI data for five subjects and ratings/features (category, dimension, visual object, and semantic) are available at figshare.
• The preprocessed fMRI data are saved as the BrainDecoderToolbox2 format.
• The raw fMRI data (bids format) is available at OpenNeuro.

Video stimuli

• We used 2185 emotion evocative short videos collected by Cowen and Keltner (2017).
• You can request the videos with emotion ratings from the following URL (https://goo.gl/forms/XErJw9sBeyuOyp5Q2).

Code

Matlab and Python scripts for reproducing main results of our study are available from code/.

The main code (emotion2020_analysis_BATCH.m) includes

• Data preparations for delineating individual ROIs, including WholeBrain, HCP360 ROIs, subcortical regions
• (regularized) linear regression analyses (encoding/decoding) between MRI data and labels (category, dimension, vision, semantic) associated with 2196 (2181 unique) emotion evocative movie clips.
• Representational similarity analysis
• K-means clustering using emotion-related brain activity (encoding results are necessary)

Preparations (set following data in the appropriate locations):

• Preprocessed fmri data (e.g., ./data/fmri/Subject1/preprocessed/fmri_Subject1_Session1.h5)
• ROI information file (./data/fmri/misc/roiInf.mat)
• Feature data (e.g., ./data/feature/category.mat)
• Principal gradient data (e.g., ./data/fmri/Subject1/pringrad/Subject1_pringrad_values.mat)
• BrainDecoderToolbox2 (./code/libraries/BrainDecodeeerToolbox2)

Main parts:

• To go through all analyses and get all results, run this and python scripts as below.

  1. run this script with setting 1 for roiDataPreparation variable (required ~20GB; multiple cpu can work in parallel; this can be skipped if you have downloaded and set ROI-wise .mat files in e.g., ./data/fmri/Subject1/rois/)
  2. run this script with setting 1 for performDecAnalyses/performEncAnalyses/performRSAnalyses variables (multiple cpu can work in parallel)
  3. run this script with setting 1 for summaryDecAnalyses/summaryEncAnalyses/summaryRSAnalyses variables
  4. run this script with setting 1 for performeAdditionalAnalysis variable
  5. run python scripts to perform UMAP analyses.
  6. run this script with setting 1 for showDecResults/showEncResults/showRSAResults variable

Note:

• Decoding, encoding, and representational similarity analyses can be performed independently.
• Two umap analysis implemented in python scripts requires results of decoding and encoding analyses.
• The analysis part (ii) will take about 1 day using 100 cpu to complete all the computations.