SpaVCCA Identifies Spatial Transcriptomics Domains Across Slices by Coupling Variational Autoencoder with Canonical...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics offers gene expression insights within tissue context.
• Technical variability and batch effects hinder data integration across studies.
• Existing methods struggle to balance batch correction with spatial structure preservation.

Purpose of the Study:

• To develop a unified framework, SpaVCCA, for robust spatial transcriptomics data integration.
• To address the challenge of identifying biological domains across diverse datasets and platforms.
• To improve the characterization of spatially organized cellular states.

Main Methods:

• SpaVCCA utilizes a graph convolutional Variational Autoencoder (VAE).
• It incorporates Canonical Correlation Analysis (CCA)-based alignment loss for batch correction.
• A graph contrastive objective is employed to preserve local spatial neighborhood structures.

Main Results:

• SpaVCCA demonstrates superior performance compared to state-of-the-art methods.
• The framework effectively integrates data from diverse platforms (Visium, Stereoseq, MERFISH) and dimensions (2D, 3D).
• It achieves significant advancements in batch correction and spatial structure preservation.

Conclusions:

• SpaVCCA offers a scalable and effective solution for integrating spatial transcriptomics data.
• The method facilitates the characterization of spatially organized cellular states.
• It supports applications in drug development and disease research by enabling robust data integration.