mackelab/sbi-for-connectomics-data

Dataset Card: SBI for Connectomics — Data and Results

This dataset contains the binary data, inference results, and figures accompanying the paper Simulation-based inference for efficient identification of generative models in connectomics.

It is the companion data repository for the code at github.com/mackelab/sbi-for-connectomics, which uses simulation-based inference (SBI) to infer parameters of computational wiring-rule models of the rat barrel cortex used in connectomics research.

These files were previously distributed via git-lfs in the code repository and have been moved here to keep the git repository lightweight.

Dataset structure

~3.0 GB across 219 binary files, organized into three top-level folders that mirror the paths expected by the code and notebooks in the companion repository:

data/
├── structural_model/       # masks and feature tables for the structural (barrel cortex) model
├── subcellular_features/   # masks and feature tables for the subcellular model
├── cube_model/             # masks, feature tables, and depth-profile data for the cube model
└── presimulated_dso_*.p    # presimulated prior/posterior samples for various wiring-rule parameterizations

figures/
├── fig2_structural_model.pdf
├── fig3/                   # main-text figure 3 (SBI results) source data/renders
├── fig4-5/                  # main-text figures 4-5 (distance-rule results)
└── supplementary_figures/   # supplementary figure renders

results/
├── amortized_posterior_*.p          # amortized NPE posteriors
├── npe_*.p                          # NPE posteriors trained on various wiring-rule datasets
├── prior_predictive_*.p / posterior_predictive_*.p
├── sbc_results_*.p                  # simulation-based calibration results
├── cross_validation_results_*.p
└── tutorial_posterior_*.p           # posteriors used in the tutorial notebooks

File formats:

• .p — Python pickle files (mostly torch/sbi posterior objects and sample arrays; see the companion repository's consbi package and notebooks for how they're loaded)
• .txt / .tsv — mask and feature tables for the structural/subcellular/cube models
• .pdf / .png / .eps — rendered paper figures

Code, notebooks, and small config/log files (plotting scripts, .ipynb notebooks, .hydra run configs, etc.) are not included here — those remain version-controlled in the GitHub repository.

How to use

Download directly into a local clone of the companion repository so the folder layout matches what the code expects:

pip install -U huggingface_hub
hf download mackelab/sbi-for-connectomics-data --repo-type dataset --local-dir .

This populates data/, figures/, and results/ in the current directory.

Citation

If you use this data, please cite the associated paper:

Simulation-based inference for efficient identification of generative models in connectomics. https://www.biorxiv.org/content/10.1101/2023.01.31.526269v1

License

MIT License, matching the companion code repository.