odor_mix_code

Analysis code and bundled source data for Sparse input representations explain odor discrimination in complex, concentration-varying mixtures.

Generated figures and newly generated result tables are ignored by default.

Repository Layout

• odor_mix_code/: installable Python package with reusable model code and project path helpers.
• scripts/: notebooks used to generate manuscript figures, plus a compatibility import.
• data/: bundled source data from Zak 2020, Burton 2022, and Zak 2024.
• results/GLUE/: precomputed GLUE result table used when the optional GLUE package is unavailable.
• figures/: generated locally by notebooks; not tracked.
• results/: generated result tables; only the bundled GLUE fallback CSV is tracked.

Setup

Create the mamba environment, install the local package, and register a notebook kernel:

mamba env create -f environment.yml
mamba activate sparse-inputs
pip install -e .
python -m ipykernel install --user --name sparse-inputs --display-name "Python (sparse-inputs)"

The notebooks write figures under figures/ and generated result tables under results/. These directories are created as needed and are ignored by Git except for the tracked GLUE fallback result.

Reproducing Figures

Run notebooks from the repository root or from the scripts/ directory. The notebooks use package path helpers, so they should not depend on a machine-specific absolute path.

Notebook	Manuscript output	Notes
scripts/1-all_decoding.ipynb	Fig. 3B-F, Fig. S5	Main decoding simulations. Some cells are longer-running because they sweep concentrations, backgrounds, and replicates.
scripts/2-glue_analysis.ipynb	Fig. 4	Can plot from results/GLUE/glomerular_poisson_gcmc_withshuffle_24bg_high_low_scale_0.0_10000points.csv if the optional gcmc package is unavailable.
scripts/3-OSN_PiC_data.ipynb	Fig. S7B	One-vs-rest logistic regression on Zak 2024 OSN/PiC data and response sparsity analysis.
scripts/4-sparse_expansive_code.ipynb	Fig. S7A	Sparse expansive code comparison. Can load a prior generated result table if available.
scripts/5-glom_data_sparsity.ipynb	Fig. 3A, Fig. S6	Glomerular model and data sparsity analysis using bundled Zak 2020 and Burton 2022 data.

See scripts/README.md for the recommended run order, inputs, and outputs.

Data

• data/Zak_2020/Glomerular_Matrix.mat: glomerular response data used in sparsity analyses.
• data/Burton_2022/Fig1figsupp3source data 1_figS3_data.mat: source data used in sparsity analyses.
• data/Zak_2024/: OSN and bouton response tables plus odor index used for OSN/PiC comparisons.
• results/GLUE/glomerular_poisson_gcmc_withshuffle_24bg_high_low_scale_0.0_10000points.csv: precomputed GLUE result table for reproducing Fig. 4 without running the optional GLUE pipeline.

Reproducibility Notes

• Model initialization uses explicit seeds in the notebooks and helper functions.
• The installable module is odor_mix_code.coding_model_fanofactor.
• Project-relative paths are exposed from odor_mix_code.paths.
• Generated figures and non-bundled result tables are intentionally ignored to keep version control focused on source inputs and analysis code.