spateo-full-pipeline

name	spateo-full-pipeline
description	Run the complete Spateo pipeline — IO through Morphogenesis

Full Pipeline Orchestrator

Run all Spateo stages sequentially: IO → Alignment → 3D Reconstruction → Morphogenesis.

Quick Start

# Run full pipeline with default settings
python -m skills.00_pipeline.run_all \
  --data-path ./data/xenium_outs/ \
  --platform xenium \
  --stage2-path ./data/stage2.h5ad \
  --out-dir ./output/

# Run with custom alignment
python -m skills.00_pipeline.run_all \
  --data-path ./data/ \
  --auto-detect \
  --slices-dir ./data/slices/ \
  --out-dir ./output/ \
  --skip-morphogenesis

Pipeline Stages

Stage 0 (IO) ──→ Stage 1 (Alignment) ─→ Stage 2 (Reconstruction)
                                                    ↓
                                     interpolation    morphology
                                                    ↓
                                        Stage 3 (Morphogenesis)
                                        vectorfield ──→ feature

Stage	What it does	Output
0: IO	Read raw spatial data from platform	adata.h5ad with .obsm["spatial"]
1: Alignment	Register slices into common coordinate space	aligned.h5ad with .obsm["align_spatial"]
2: Reconstruction	Build 3D models + interpolate expression	VTK models, interpolated data
3: Morphogenesis	Infer vector fields + compute features	Vector field layers, DEG tables

CLI Parameters

Parameter	Purpose	Default
--data-path	Raw data directory (Stage 0)	—
--platform	Platform name (e.g. xenium, merfish)	--auto-detect to guess
--slices-dir	Directory with slice .h5ad files (Stage 1)	Uses IO output
--stage2-path	Second stage data for morphogenesis (Stage 3)	—
--out-dir	Output directory for all stages	./output/
--skip-alignment	Skip Stage 1	—
--skip-reconstruction	Skip Stage 2	—
--skip-morphogenesis	Skip Stage 3	—
--model-type	3D reconstruction model type	surface
--interpolation-method	Gene interpolation method	vtk
--device	Compute device	Auto-detect
--verbose	Enable debug logging	—

Programmatic Usage

from skills.00_pipeline.run_all import PipelineConfig, run_full_pipeline

config = PipelineConfig(
    data_path="./data/xenium_outs/",
    platform="xenium",
    stage2_path="./data/stage2.h5ad",
    out_dir="./output/",
)
result = run_full_pipeline(config)

Troubleshooting

Problem	Fix
Stage 0 fails	Check data directory contains platform-specific files
Stage 1 fails	Verify .obsm["spatial"] exists in input AnnData
Stage 2 OOM	Reduce voxel_size or use --model-type pointcloud
Stage 3 fails	Ensure stage1 and stage2 have matching gene sets
Any stage hangs	Set --device cpu to avoid GPU issues