# Data Repository This repository contains the subject-level data that was used in Jagadeesh & Gardner, PNAS 2022 (https://www.pnas.org/doi/full/10.1073/pnas.2115302119). These data are already preprocessed using the GLMdenoise package (Kay et al., 2013), so what is found here are condition-level betas, i.e. the average response of each voxel to each unique image after regressing out noise. There were 30 unique images (conditions) presented to the subject: 10 different image classes x 3 types (original, synth1, synth2). ## Description of relevant fields in each subject's .mat file: * "amplitudes": is a cell with two arrays: the first array corresponds to the left hemisphere voxels and the second corresponds to the right hemisphere voxels. The dimensions of these arrays are voxels x conditions. So in this case there are 4325 voxels in the left hemisphere and 4877 voxels in the right hemisphere, and there are 30 conditions. * "whichRois": is also a cell containing two arrays. In this case, the dimensions of each array are just voxels x 1, and each voxel is labeled with a numerical label which corresponds to an element of the roiNames field. * "roiNames": is a 1x30 cell array of strings where each element is the name of an ROI. * "stimNames": is a 1x30 cell array of strings indicating the condition. There are 30 total conditions: 3 different sample values x 10 different image names. Sample can be either -1 (original), 1 (synth sample 1), or 2 (synth sample 2). Image name can be 1 through 10. Each numerical value of image name corresponds to an image whose name can be found in the imageNames field. * "stimValues": is a 2x30 array that also indicates the condition name. In this case the first row indicates the sample value in each of the 30 conditions (either -1, 1, or 2), and the second row indicates the image name (1 through 10). Please don't hesitate to reach out to the first author, Akshay Jagadeesh, if you have any questions about this paper or these data.