Welcome to SECE’s documentation!

Introduction:

SECE is designed to learning effective low-dimensional features for ST-seq data. It is a framework for spatial region-related embedding (SE) learning and cell type-related embedding (CE) learning.

SECE has two modules:

• AE Module: Autoencoder with negative binomial distribution to model expression counts and learn CE.

• GAT Modlue: Graph Attention network to learn SE using adjacency matrix and similarity matrix constructed from spatial location.

By applying SECE to diverse ST-seq data with different resolutions and different tissue types, we obtained state-of-the-art spatial domain identification results and demonstrated that SE can be used for tasks such as visualization and trajectory inference. Here are examples of the installation and use of SECE.

Usage:

• Installation
• 1. Stereo-seq Olfactory bulb
• 2. Stereo-seq Hemibrian
• 3. Slide-seqV2 Hippocampus
• 4. STARmap Visual cortex
• 5. Visium Breast cancer

Citation: