Higher-level spatial prediction in natural vision across mouse visual cortex

This repository contains the complete analysis code and computational resources to reproduce all results from "Higher-level spatial prediction in natural vision across mouse visual cortex" by Heilbron & de Lange. Importantly, this repository does NOT contain a novel dataset. All neural data analysed in this study comes from the publicly available Allen Institute Visual Coding Neuropixels dataset (Siegle et al., 2021), and no neural or stimulus data is redistributed. Instead, we provide automated scripts to download data directly from the Allen Institute Brain Observatory API and organize it into analysis-ready formats as described in the Methods. The repository includes complete Python implementations of all analyses presented in the paper. To facilitate reproducibility and reduce computational burden, we provide pre-computed spatial predictability estimates for all neurons and stimuli, multi-level unpredictability metrics across abstraction levels, low-level image statistics, and all regression model results. Additionally, we include fine-tuned model checkpoints of the deep neural networks. Together, the code, models and numerical results, provide everything to reproduce the analyses and figures in the paper.