Quantification of pan-expansion-microscopy (pan-ExM) brain volumes: synapses and the expansion factor from the DP–PSD distance, nuclei as an independent ExF ruler, multi-antibody organisation at synapses, and QC for light-sheet (pan-ASLM) data. Python, works from confocal stacks (OME-Zarr) to terabyte Neuroglancer-precomputed volumes.
The method follows M'Saad, Cairns et al., Nature Biotechnology 2025 (doi:10.1038/s41587-025-02905-4); what was taken from the paper and what was written here is set out in legacy/ExF_Quantification/METHODS_AND_CREDITS.md.
| Folder | What |
|---|---|
| ebq/ | The package. 3D synapse finder, nucleus finder, the profile measurement, a dataset manifest (channel roles, animal, genotype, sex), microscope profiles, readers for precomputed and OME-Zarr. See docs/ebq_package.md. |
| tests/ | Synthetic-data tests (python -m pytest). |
| docs/ | Working notes on the server layout, Neuroglancer/OME-Zarr gotchas and analysis decisions (docs/README.md); the package overview. |
| legacy/ | The Windows tools used in 2026: OME-Zarr batch converter, image list with Neuroglancer links, QC screenshots, gallery, and the original 2D synapse/ExF analysis. Still runnable; superseded by ebq for new analysis. See legacy/README.md. |
pip install -e .
ebq manifest-template datasets.csv # one row per dataset: path, animal, genotype, sex, channel roles
ebq manifest-check datasets.csv
ebq synapses3d <precomputed folder> --out out/syn --fields 20
ebq nuclei <precomputed folder> --out out/nuclei --exf 20
Requires Python 3.10+, numpy ≥ 2, scipy, scikit-image, zarr ≥ 3, dask, matplotlib (installed by pip install -e .).
ebq.synapse3d: on a pan-ASLM volume it recovers 9 of 12 synapses logged by the 2D tool and gives ExF 23.7 ± 1.6 across fields where the 2D tool gave 21.7 ± 1.6. Not yet validated against hand measurements.ebq.nuclei: finds complete nuclei in a whole volume; recall on light-sheet data is still low (large nuclei merge with neighbours).- The expansion factor of the pan-ASLM volumes must be measured on the gel; the DP–PSD method only reports values inside its 12–30× search window.
- Mitochondria: use MAPS (Tian et al. 2025) on the 60× confocal pan-stain stacks; the pretrained kidney model works on brain tissue out of the box.
pan-aslm2neuroglancer — pan-ASLM tiles → stitched Neuroglancer precomputed on the HPC. kuan-lab/pan-exm-t_paper — the paper's conversion code.