SCALPEL Pipeline

---

This repository contains the SCALPEL (Spatial Cell Analysis, Labeling, Processing, and Expression Linking) processing pipeline for spatial transcriptomics data analysis.

• Preprint:https://www.biorxiv.org/content/10.64898/2026.01.09.698732v1
• Repository:https://github.com/AllenInstitute/Spatial-Transcriptomics-Processing-Pipeline

This pipeline was built and deployed on Code Ocean, a cloud-based computational research platform. The pipeline leverages Code Ocean's containerized environment to ensure reproducible results across different computing environments.

Code Ocean documentation: https://docs.codeocean.com/user-guide

QC & Mapping Overview

The processing section of pipeline consists of the following sequential steps:

Step 1: QC Filtering & Doublet Detection

Quality control filtering and doublet detection with SOLO (Semi-supervised Outlier Detection).

Step 2: Cell Type Mapping

Perform MapMyCells cell type mapping.

Step 3: Combine Sections

Aggregate individual section-level AnnData files into a single combined dataset for whole-dataset analysis.

Step 4: Add Cell Type Colors

Add color mappings for cell type classifications to AnnData objects using the ABC atlas color scheme.

Step 5: DoubleMAD Mapping Filtering

Perform quality control on cell type mapping results using Double Median Absolute Deviation (DoubleMAD) statistics to identify and filter cells with poor mapping confidence scores.

Step 6: Save Processing Results

Save final processed results from the pipeline and add final QC column.

Domain Detection Overview

The domain detection section of pipeline consists of the following sequential steps:

Step 1: Downsample Spot Table

Bin transcript spots and performs QC filtering.

Step 2: Run STAligner

Perform spatial alignment and integration of multiple tissue sections using STAligner.

Step 3: Leiden Clustering

Perform Leiden clustering with STAligner embeddings.

Step 4: Add Clusters to Cell-By-Gene

Map cluster assignments from downsampled STAligner gridded data to cell segmentation data.

Step 5: Merge All Clusters

Consolidate cluster assignments to full processed dataset.

Setup

Running via Code Ocean UI:

1. Create a new Pipeline by cloning this repository
2. Replace Data Parameters
3. Configure App Panel with your dataset-specific parameters
4. Verify data format matches expected input structure
5. Click "Run with parameters" to run pipeline

Running on your local machine:

1. Click Pipeline -> Export
2. Follow the instructions in REPRODUCING.md

Configuration

All pipeline parameters are configured in the Create Parameters JSON capsule and centralized in params.json. Key parameter categories include:

• QC Filtering Parameters
• Mapping Parameters
• Metadata Parameters
• Domain Detection Parameters

Input Data Format

data/
└── section_dir/
├── section1.h5ad
├── section2.h5ad
├── section3.h5ad
└── ...
...

Required columns:

• x and y: cell centroid coordinates
• brain_section_barcode: Section ID
• Index containing unique cell labels (e.g.,{brain_section_barcode}_SIS_{i})

Output Files

The pipeline generates the following key outputs:

results/
├── whole_dataset/
│ ├── {specimen}_{dataset_id}_filtered.h5ad
| └── {specimen}_{dataset_id}_filtered.csv
└── sections/
├── section1_filtered.h5ad
├── section2_filtered.h5ad
└── ...

Key output files:

• {specimen}_{dataset_id}_filtered.h5ad: Combined, QC-filtered data
• sectioni_filtered.h5ad: QC-filtered data split by brain_section_barcode

Support

For detailed information about each step, refer to the individual markdown files linked above. Each file contains:

• Detailed methodology description
• Input/output file specifications
• Configuration parameter explanations
• Expected results and metadata columns

We are planning on occasional updating this tool with no fixed schedule. Community involvement is encouraged through both issues and pull requests.