stereo-seq-3d-reconstruction

name: stereo-seq-3d-reconstruction description: Use when Stereo-seq or STOmics serial sections need multi-slice registration, 3D coordinate reconstruction, slice-to-atlas alignment, shape/contour matching, SABench-informed alignment method choice, or 3D-ready visualization and mapping.

Stereo-seq 3D Reconstruction

Use This For

• Aligning serial Stereo-seq slices or histology masks into a shared coordinate frame.
• Mapping spatial coordinates through thin-plate-spline or other slice-registration transforms.
• Preparing registered coordinates for 3D reconstruction, atlas projection, or cross-section comparison.
• Selecting or auditing 3D spatial alignment methods using the Nature Computational Science 2026 SABench benchmark guidance.

Default Requirements

• Use bundled article-derived scripts in scripts/ before writing new registration or plotting code.
• Read source_code.md first for curated templates and SABench alignment-method guidance; if no curated entry fits, search code_candidates.tsv for additional article-linked repositories and reusable files.
• Inspect local Python/R environments first. Prefer conda run -n stereo-skills-py python ... for Python scripts and conda run -n stereo-skills-r Rscript ... for R scripts. spatiAlign is configured in stereo-skills-py; it can also be isolated with envs/environment-python-spatialign.yml. If a dependency is missing, stop that step and report the missing package and blocked script.
• Use equal-aspect spatial plots, readable Arial text, non-overlapping legends, and export PDF/300 dpi diagnostic figures.
• In the final response, state the reused paper, DOI, code repository/source file, and what was changed for the current dataset.

Workflow

1. Identify section order, coordinate units, tissue masks, anchors, and whether registration is pairwise or to a shared atlas.
2. Read source_code.md, then decide whether the task is method selection/audit or production coordinate transformation.
3. For method selection/audit, use SABench guidance to document why the chosen method matches the data context: serial sections, cross-platform slices, large-scale/high-resolution data, partial overlap, rotation sensitivity, landmarks available, or 3D-coordinate ground truth available.
4. For production coordinate transformation, adapt the closest local script:
   • scripts/sabench_alignment_runner_template.py for SABench-style PASTE, SLAT, STalign, or QC-only alignment/evaluation from serial h5ad slices.
   • scripts/zebrafish_tps_slice_registration_template.py for mask/anchor-based TPS coordinate mapping and QC plots.
   • scripts/spatialign_multislice_alignment_template.py for spatiAlign-style multi-slice spatial alignment when that package is installed.
   • scripts/multislice_spatial_qc_template.py for SLAT/SPACEL-style multi-slice coordinate QC panels and 3D-ready stacked previews before or after registration.
5. Export transformed coordinates and a registration QC plot for every slice pair or atlas mapping.
6. Validate registration with landmarks, tissue contours, known anatomical layers, gene-expression similarity, and non-overlap of impossible tissue regions.
7. Keep biological interpretation separate from registration quality.

Output Expectations

• Registration method and required inputs.
• Transformed coordinate tables.
• SABench-style runtime/memory/QC/provenance tables when alignment is run through the SABench template.
• QC plots with anchors/contours and equal aspect ratio.
• Caveats for scale, orientation, missing tissue, and manual anchors.
• Reused article code source, paper DOI, repository URL, original file name, and dataset-specific edits.