Brain-Model Alignment Derivatives for Scene Perception Research

Derivative files support the reproduction of analyses from the manuscript 'Shared representations in brains and models reveal a two-route cortical organization during scene perception'. The dataset contains representational similarity analyses (RSA), centered kernel alignment (CKA) results, model-feature extractions, and stimulus metadata for four raw fMRI datasets. It is designed to enable full figure and analysis reproduction without reprocessing the underlying neuroimaging data.

Use Cases

• Reproducing representational similarity analyses (RSA) based on pre-computed neural and model feature alignments.
• Comparing brain-model alignment using centered kernel alignment (CKA) results across multiple fMRI datasets.
• Extracting language-model features for stimuli using provided human-curated and Pixtral-12B generated captions.
• Analyzing cortical organization with provided HCP-MMP1 atlas definitions and model metadata.

Strengths

• Enables full reproduction of a peer-reviewed manuscript's figures and analyses without raw data reprocessing.
• Integrates derivatives from four established fMRI datasets (NSD, BOLD5000, THINGS-fMRI, Mosaic-fMRI).
• Includes both human-curated and AI-generated (Pixtral-12B) stimulus captions for language-model feature extraction.
• Companion GitHub repository provides all analysis code and generation instructions.

Limitations

• Dataset size is large (approximately 4.6 GB), which may limit accessibility for some users.
• Specific column names and row counts are not detailed in the provided metadata.
• As derivative data, its utility is contingent on the user's familiarity with the original raw datasets and the specific analytical framework.

Provenance

Source

Derived from analyses for the manuscript by Pablo Marcos-Manchón (arXiv:2507.13941).

Collection Method

Generated as part of the peer-review revision process for the associated computational neuroscience research.

Freshness

Last updated on 2026-03-20.