Derivatives for "Shared representations in brains and models reveal a two-route cortical organization during scene perception"

This dataset contains all derivative files generated for the analyses presented in the manuscript Shared representations in brains and models reveal a two-route cortical organization during scene perception (arXiv:2507.13941). These derivatives include representational similarity analyses (RSA), centered kernel alignment (CKA), model-feature extractions, stimulus metadata, and additional materials produced during the peer-review revision process.The derivatives allow full reproduction of the figures and analyses presented in the paper without requiring reprocessing of the raw fMRI datasets (NSD, BOLD5000, THINGS-fMRI, Mosaic-fMRI). All analysis code, together with instructions for generating the derivatives from raw data, is provided in the companion repository:https://github.com/memory-formation/convergent-transformationsA full description of each file is available in derivatives/README.md within the repository.The dataset is organized into the following folders:1. metadata/Global metadata used across analyses:HCP-MMP1 atlas definitionsModel metadata and parameter counts2. captions/Stimulus descriptions used for language-model feature extraction:Human-curated MS-COCO captionsPixtral-12B generated captions for NSD, BOLD5000, and THINGS-fMRI3. nsd/Representational alignment derivatives for the Natural Scenes Dataset:Inter-subject RSA and CKA (multiple shifts and hemisphere configurations)Subject–model RSA for all vision and language modelsPartial RSA controlling for confoundsPermutation distributions for significance testingkMCCA projections and semantic dimensionsScene/object annotations4. bold5000/Derivatives for the BOLD5000 dataset:Inter-subject RSA across all images and dataset subsetsSubject–model RSA across visual and language models5. things/THINGS-fMRI derivatives:Inter-subject RSASubject–model RSA6. datasets/Preprocessed metadata supporting all alignment analyses:NSD stimulus indices, ROI masks, and image filesBOLD5000 stimulus indices, captions, and trial mappingsTHINGS-fMRI categories, captions, and stimulus indices7. revision/Additional files generated during the peer-review revision process:ROI voxel countsReliability analyses (stepwise subset analysis, half-split, repetition-based)MPNet (sentence-transformers) representations following Doerig et al. 2025How to use this dataset:Place these derivatives inside a derivatives/ directory at the root of the GitHub repository. All notebooks and scripts will run without recomputing intermediate files.Code availability:All code and instructions are provided in the public repository:https://github.com/memory-formation/shared-representationsContact:For any questions or additional materials, contact the corresponding author (Pablo Marcos-Manchón; [e-mail]).