mcDETECT tutorial: MERSCOPE analysis (step by step)

This tutorial follows the repository workflow for MERSCOPE data: transcript-based mcDETECT setup → granule detection → expression profiling → manual granule subtyping → WT vs AD granule density comparison → Isocortex neuropil subdomain analysis on a 50×50 grid with hard embeddings, row-normalized subtype vectors, and K-Means with k = 4.

Primary references: mcDETECT_package/mcDETECT/model.py, code/3_detection.py, code/benchmark/benchmark_subtyping.ipynb, code/7_neuropil_subdomains.ipynb.

Analysis roadmap

  1. Prepare — transcript and auxiliary tables (coordinates, gene names, nucleus overlap, optional cell_id).

  2. Initialize mcDETECT with platform type, marker lists, and numerical hyperparameters.

  3. Detect granules — optional rough pass, then fine pass with filters.

  4. Profile granules into an AnnData count matrix.

  5. Subtype manually — normalize expression, k-means, heatmap, map clusters to biology, derive a simple mixed/pure label.

  6. Compare density WT vs AD by brain region using spots as spatial units.

  7. Neuropil subdomains (Isocortex, 50×50) — hard spot_embedding, row-wise normalization, K-Means k = 4.

See also the quick reference checklist.

Dataset-specific coordinate transforms and multi-setting benchmark loops are omitted here.

Pages

File

Section

01_what_mcdetect_expects.md

What mcDETECT expects

02_construct_mcdetect.md

Step A — Construct mcDETECT

03_granule_detection.md

Step B — Granule detection

04_profile_granules.md

Step C — Profile granules

05_manual_subtyping.md

Step D — Manual granule subtyping

06_density_wt_ad.md

Step E — WT vs AD density

07_neuropil_subdomains.md

Step F — Neuropil subdomains (Isocortex)

08_checklist.md

Quick reference checklist

Contributors: publishing workflow for this site (not part of the RTD build): maintainer/PUBLISHING_READTHEDOCS.md in the mcDETECT-tutorial repository.