Run MEA-seqX

Runs the paper's own official demo dataset through the vendored code, essentially unmodified. Step 1 (Multiscale Spatial Alignment.py's Multiscale_Spatial_Alignment()) and Step 2 (SRT Gene Expression.py's gene_expression()) are called directly — real spaceranger capture, real BrainWave-processed electrophysiology, real human-curated reference points, all from Zenodo and the repo's own bundled data/SD/ files. See the overview for the exact data provenance and the handful of real bugs that had to be fixed to get this 2021-era code running on a modern Python stack.

Gene list (from the paper's own gene_list_all.xlsx)

Before alignment — each dataset on its own

Real SRT spots on the real tissue photo, before any registration — the Zenodo SRT record ships several real reference images (this is tissue_lowres_image.png). The 5 gold stars are the authors' own real, human-picked landmarks (

data/SD/SRT Reference
          Points.csv

) — the same 5 points shown on the nEphys grid to the right.

Real electrode grid, its own native chip (row, col) coordinates, before any registration. No reference photo exists for this — the official nEphys Zenodo record (10.5281/zenodo.10620559) contains only .brw/.bxr binary recordings, confirmed by listing its files; there's no chip or slice photo bundled with it to overlay on. The 5 gold stars are the same real landmarks as the SRT side — only 2 ("Distal CA1", "Proximal CA3") are actually used to compute the registration; the rest are shown for reference.

Step 1 — electrodes + SRT spots, one shared frame of reference

Real electrode positions (after the real move+rotate registration) and real SRT spot positions, uniformly colored (blue/orange) so shape and position can be judged on their own. Gold/lime stars mark the same 5 real landmarks shown on the "before" images above, now in this shared frame — the µm labels are the real distance between each landmark's SRT position and where it actually landed after registration. "Distal CA1" and "Proximal CA3" are the only two points the whole transform is fit from, so both land near-exactly (0µm) by construction; the other 3 are genuinely held-out and their real residual (~170–780µm here) is an honest measure of how well a 2-point rigid registration generalizes beyond the two points it was fit to — see the overview for a real rotation-angle bug this project found and fixed in the vendored code (confirmed by this same check: before the fix, Proximal CA3 itself was 722µm off instead of 0).

Step 1 — full diagnostic figure (vendored code, unmodified)

This is the paper's own real reference-point registration, run through Multiscale_Spatial_Alignment() exactly as published (aside from a few real Python 2021→now compatibility bugs fixed along the way — see the overview page). "Distal CA1" and "Proximal CA3" are the authors' own manually-picked landmarks, bundled with the repo, matched to the real barcodes and electrodes in this exact dataset. The photo above is the real H&E tissue image with every real in-tissue spot plotted at its real position, colored by real cluster — solid dots got a real electrode registered underneath them by Step 1, faint dots didn't. The table below it is the same result as data: barcode, position, cluster, and the real electrode channel IDs.

nEphys event detection & denoising (real BrainWave output, real vendored denoising)

Cluster correlation

Cluster	Events/min/channel	Mean expression	Channels