Data from: Decoding predicted future states from the brain’s ‘physics engine’
Data files
Sep 28, 2026 version files 1.47 GB
-
expt1_data.mat
166.40 MB
-
expt2_data.mat
1.30 GB
-
figure2_ContactVsNoncontact_analysis.m
5.16 KB
-
figure2_ContainVsOcclude_analysis.m
4.79 KB
-
figure4_contact_prediction.m
9.58 KB
-
figureS2_contactType_decoding_analysis.m
4.17 KB
-
README.md
4.97 KB
Abstract
Successful engagement with the physical world requires the ability to predict future events and plan interventions to alter that future. Growing evidence implicates a set of regions in the human parietal and frontal lobes in such intuitive physical inferences (aka the ‘Physics Network’ or PN). But the central tenet of this hypothesis, that the PN runs forward simulations to predict future states, remains untested. In this pre-registered study, we first show that the PN abstractly represents whether two objects are in contact with each other, a physical scene property critical for prediction (since objects' fates are intertwined when they are in contact). We then show that the PN (but not other visual areas) carries abstract information about predicted future contact events (i.e., collisions). These findings support the hypothesis that the PN contains a generative model of the physical world that conducts forward simulations, serving as the brain's "Physics Engine".
Dataset DOI: 10.5061/dryad.2547d7x35
Description of the data and file structure
This repository contains the data from the two fMRI experiments (expt1_data.mat and expt2_data.mat) and MATLAB scripts to generate plots from Figure 2/S2 (figure2_ContactVsNoncontact_analysis.m, figure2_ContainVsOcclude_analysis.m, and figureS2_contactType_decoding_analysis.m) and Figure 4 (figure4_contact_prediction.m) in the paper "Decoding predicted future states from the brain’s ‘physics engine’".
Files and variables
File: figure2_ContainVsOcclude_analysis.m
Description: MATLAB script to generate plot in Figure 2C. Specifically, it computes the average decoding index for contain vs occlude relationships in the four functional regions of interest (fROIs) across all participants.
File: figure2_ContactVsNoncontact_analysis.m
Description: MATLAB script to generate plot in Figure 2B. Specifically, it computes the average decoding index for contact vs. non-contact relationships in the four functional regions of interest (fROIs) across all participants.
File: figure4_contact_prediction.m
Description: MATLAB script to generate plots in Figure 4B and 4D. It computes the average decoding index for contact vs. non-contact decoding across predicted and perceived scenarios in the four functional regions of interest (fROIs) across all participants.
File: figureS2_contactType_decoding_analysis.m
Description: MATLAB script to generate plot in Figure S2
File: expt2_data.mat
Description: Contains data from Experiment 2.
Variables
- subjname is a cell array containing the subject IDs of the 14 participants scanned for this fMRI experiment.
- conditions is a structure containing two fields
- obs: for the observed or perceived conditions. Contains binary arrays 'isroll' and 'ishit' indicating whether each of the 48 stimulus videos used in this condition belonged to the Roll/Throw scenarios or whether the agent object hit/miss the patient object (i.e., collision or non-collision).
- pred: similar to **obs **but for the predicted conditions.
- fROIs is a structure containing voxel IDs corresponding to specific fROIs in each participant
- V1/LOC/VTC/PN: cell array of size nsubjects x 2 (Left or Right hemisphere) containing voxel IDs for the corresponding fROI in the betas variable. For example, fROIs.V1{4,2} has voxel IDs for the right hemisphere of V1 in the fourth participant. Use these voxel IDs to pick rows in the betas variable of the corresponding participant to get voxel activations for various experimental conditions (in the conditions variable). The voxels are chosen at p <0.001 (or p<0.05, see Methods in the paper) in an independent localizer.
- betas is a structure containing obs and pred fields corresponding to the observed/perceived and predicted conditions respectively. Each field is an nsubjects-length cell array containing whole-brain beta activations for all 48 video stimuli. Each cell is a participant with dimensions nvoxels x nstimuli
File: expt1_data.mat
Description: Contains data from Experiment 1.
Variables
- subjname is a cell array containing the subject IDs of the 14 participants scanned for this fMRI experiment.
- conditions is a structure containing two fields
- scenes: with further subfields for 'create', 'consequence/conseq' and 'blender' scenarios. Each subfield is an array containing the column IDs in each cell of the **'betas' **variable.
- relations: with further subfields for 'contain', 'support', 'attach', and 'occlude' relationships. Each subfield is an array containing the column IDs in each cell of the **'betas' **variable.
- fROIs is a structure containing voxel IDs corresponding to specific fROIs in each participant
- V1/LOC/VTC/PN: cell array of size nsubjects x 2 (Left or Right hemisphere) containing voxel IDs for the corresponding fROI in the **betas **variable. For example, fROIs.V1{4,2} has voxel IDs for the right hemisphere of V1 in the fourth participant. Use these voxel IDs to pick rows in the betas variable of the corresponding participant to get voxel activations for various experimental conditions (in the conditions variable). The voxels are chosen at p <0.001 (or p<0.05, see Methods in the paper) in an independent localizer.
- betas is a nsubjects-length cell array containing whole-brain beta activations for all 12 experimental conditions. Each cell is a participant with dimensions nvoxels x nconditions
Code/software
Tested on MATLAB R2018b
Human subjects data
The fMRI data has been de-identified and presented in a matrix (MATLAB .mat) format. The participants gave consent to a standard protocol from MIT COUHES.