Coupling between dendritic arborization and intrinsic membrane physiology in the released mouse and human reconstructions of the Allen Cell Types Database

Reproducible Python code for a secondary analysis of publicly released mouse and human cortical neuron morphology and intrinsic electrophysiology from the Allen Cell Types Database.

Overview

The Allen Cell Types Database released standardized whole-cell recordings together with three-dimensional neuron reconstructions for 2,336 mouse and human cortical specimens, making a whole-release, cross-modal analysis possible without new tissue. This project re-derives arborization morphometrics for all 701 released SWC reconstructions — adding axon-level geometry, per-cell 3-D Sholl profiles, and the apical/basal compartment split that the released summary tables omit — and links morphology to intrinsic physiology cell-by-cell. All inference treats the donor, not the cell, as the sampling unit (delete-one-donor jackknife, donor-cluster bootstrap). The analyses quantify morphology-physiology coupling in both species, decompose how much of the species contrast reflects cortical area, sample composition, and morphological range, and stress-test what morphology predicts under comparable sampling distributions.

Main analyses

• Cohort construction from the released metadata tables (species, class, layer, area, electrophysiology QC, human donor metadata from public release fields)
• SWC morphometric re-derivation with parsing-consistency validation against the released summaries
• 3-D Sholl analysis and an apical/basal compartment split from SWC type codes
• Morphology-physiology coupling matrices with donor-aware inference and BH-FDR control (150 feature pairs per species)
• Class signatures, laminar gradients, and cortical-area controls
• Restriction ladder, direct residual contrasts, and matched-sample designs for the species coupling contrast
• Reconstruction-completeness, redundancy, range-matching, and common-support sensitivity analyses
• Donor-grouped (leave-one-donor-out) ridge prediction benchmark with compartment, common-support, range-matched, and composition stress tests

Repository structure

allen-morphophysiology/
├── README.md
├── LICENSE
├── CITATION.cff
├── requirements.txt
├── .gitignore
├── code/
│   ├── build_cohort.py          # metadata tables -> cohort annotations, QC, donor table
│   ├── morphology.py            # SWC parsing, morphometrics, Sholl, completeness proxies
│   ├── build_analysis_table.py  # morphology x electrophysiology x donor join
│   ├── statistics.py            # donor-aware coupling, contrasts, ladder, class/laminar analyses
│   ├── sensitivity.py           # completeness, range-matching, redundancy, CCA
│   └── prediction.py            # donor-grouped ridge benchmark and stress tests
└── docs/
    ├── DATA.md                  # how to obtain and arrange the source data
    └── METHODS.md               # concise methods companion

Data

Source data are not included. The pipeline reads the public Allen Cell Types Database metadata tables and SWC reconstruction files that you retrieve yourself from the Allen Brain Map API; docs/DATA.md documents the exact components, API query patterns, expected column names, and the local data/ layout the scripts expect. The snapshot used in the manuscript was accessed 2026-09-17.

Installation

python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -r requirements.txt

Python 3.12 was used for the published analyses; the code needs only NumPy, SciPy, pandas, and scikit-learn.

Reproducing the analyses

Run the modules in order from the repository root (each also accepts --data-dir / --results-dir):

python code/build_cohort.py          # cohort tables from the metadata export
python code/morphology.py            # SWC morphometrics (all 701 reconstructions)
python code/build_analysis_table.py  # analysis-ready join
python code/statistics.py            # primary coupling, contrasts, ladder
python code/sensitivity.py           # robustness and sensitivity analyses
python code/prediction.py            # prediction benchmark and stress tests

code/statistics.py, code/sensitivity.py, and code/prediction.py share their resampling primitives (donor-aware jackknife and cluster bootstrap, BH-FDR, distribution matching) through code/statistics.py. All resampling is seeded (bootstrap 42, matching 43, prediction partitions 999 + repetition), so runs are deterministic. Outputs are written under results/.

Citation

See CITATION.cff. The v1.0.0 release is archived on Zenodo: DOI 10.5281/zenodo.23135429. Please cite both the repository archive and the accompanying manuscript.

License

MIT — covers the code only. The Allen Cell Types Database source data are not redistributed here and remain governed by the Allen Institute's own terms (see docs/DATA.md).

Paper

This repository accompanies the manuscript "Coupling between dendritic arborization and intrinsic membrane physiology in the released mouse and human reconstructions of the Allen Cell Types Database."