SEPAR Tutorial

Contents

• Tutorial 1: A Detailed Quick Start Guide (DLPFC dataset)
• Tutorial 2: mouse olfactory bulb generated by Stereo-seq
• Tutorial 3: mouse somatosensory cortex by osmFISH
• Tutorial 4: mouse embryonic (E15.5) brain by MISAR-seq data(Multi-omics Dataset)
• Tutorial 5: multi-slice analysis (DLPFC dataset)
• Tutorial 6: SEPAR Analysis on Human Colorectal Cancer Visium HD Dataset

Introduction

SEPAR (Spatial gene Expression PAttern Recognition) is a computational method designed for analyzing spatial transcriptomics data. This tutorial will guide you through the installation and basic usage of SEPAR.

Main Features

• Unsupervised spatial pattern recognition

• Pattern-specific gene identification

• Gene expression pattern refinement

• Improved spatially variable gene detection

• Spatial domain clustering

• Multi-omics Integration

SEPAR Repository

The SEPAR repository is hosted on GitHub and contains all the necessary code and documentation:

• Repository URL: https://github.com/zerovain/SEPAR

• Repository Structure:

  • SEPAR_model.py: Core implementation of the SEPAR algorithm

  • SEPARmult_model.py: multi-slice version of the SEPAR algorithm

  • example/: Example notebooks

Installation

Prerequisites

Before installing SEPAR, ensure you have Python 3.7 or later installed. SEPAR requires the following package dependencies:

pandas>=2.0.3
numpy>=1.23.5
scanpy>=1.9.6
anndata>=0.8.0
matplotlib>=3.6.1
scipy>=1.10.0
scikit-learn>=1.2.0
tqdm>=4.64.1
seaborn>=0.13.2
scikit-misc>=0.2.0

# Optional dependency
# cupy>=10.6.0

Setting Up the Environment (3 minites)

1. We recommend using conda to create a new environment:

conda create -n separ python=3.8
conda activate separ

2. Install the required packages:

conda install pandas numpy scipy matplotlib scikit-learn tqdm seaborn
conda install -c conda-forge scanpy anndata scikit-misc

Installing SEPAR (1 minute)

Clone the SEPAR repository from GitHub:

git clone https://github.com/zerovain/SEPAR.git
cd SEPAR

Quick Start

Here’s a minimal example to get you started with SEPAR:

import scanpy as sc
from SEPAR_model import SEPAR

# Load your data (example with anndata format)
adata = sc.read_h5ad('your_data.h5ad')

# Initialize SEPAR
separ = SEPAR(adata, n_cluster=8)

# Preprocess the data
separ.preprocess(min_counts=0, min_cells=0, n_top_genes=3000)

# Compute spatial graph
separ.compute_graph()

# Compute weights
separ.compute_weight()

# Run SEPAR algorithm
separ.separ_algorithm(r=30, alpha=1.0, beta=0.1, gamma=0.1)

# Perform clustering
separ.clustering()

Detailed Usage

Data Preprocessing

The preprocess function handles basic data preprocessing:

separ.preprocess(
    min_counts=0,     # Minimum counts required for a cell to pass filtering
    min_cells=0,      # Minimum cells required for a gene to pass filtering
    n_top_genes=3000, # Number of highly variable genes to select
    normalize=True    # Whether to normalize the data
)

Spatial Graph Construction

Compute the spatial neighborhood graph:

separ.compute_graph(
    rad_cutoff1=None,     # Radius cutoff for neighborhood graph
    radius_rate=1.2       # Rate to adjust the radius if radius cutoff is not given
)

Pattern Recognition

Run the SEPAR algorithm to identify spatial patterns:

separ.separ_algorithm(
    r=30,           # Number of patterns to identify
    alpha=1.0,      # Weight for graph regularization
    beta=0.1,       # Weight for sparsity penalty
    gamma=0.1,      # Weight for pattern orthogonality
)

Clustering and Visualization

Perform spatial domain clustering and visualize results:

# Clustering
separ.clustering(n_cluster=8)

Example Datasets

For demonstration purposes, you can use publicly available spatial transcriptomics datasets, such as:

• 10x Visium datasets

• Stereo-seq datasets

• osmFISH datasets

• MISAR-seq datasets

Output Interpretation

SEPAR generates several key outputs:

1. Spatial patterns (accessible via separ.Wpn)

2. Gene loadings (accessible via separ.Hpn)

3. Clustering results (accessible via separ.labelres)

4. Pattern-specific genes (can be identified using identify_pattern_specific_genes())

Citation

If you use SEPAR in your research, please cite:

Zhang, L., Zhu, Y. & Zhang, S. SEPAR enables spatial metagene discovery
and associated molecular pattern characterization in spatial transcriptomics
and multi-omics datasets. Commun Biol 9, 77 (2026).
https://doi.org/10.1038/s42003-025-09340-w

Support

For questions and issues, please visit our GitHub repository or contact us through the issues page.

License

SEPAR is released under the MIT License. See the LICENSE file in the repository for more details.