SGCAST: symmetric graph convolutional auto-encoder for scalable and accurate study of spatial transcriptomics.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) provides gene expression data with spatial context.
• High-resolution ST data presents challenges in efficient and scalable spatial domain identification.
• Accurate analysis of the tissue microenvironment is crucial for biological insights.

Purpose of the Study:

• To develop an efficient and scalable framework for identifying spatial domains in spatial transcriptomics data.
• To improve the accuracy and performance of spatial domain identification methods.
• To enable comprehensive analysis of high-resolution ST datasets.

Main Methods:

• Developed SGCAST, a symmetric graph convolutional auto-encoder framework.
• Integrated gene expression similarity and spatial spot proximity for latent embedding learning.
• Implemented a mini-batch training strategy for memory efficiency and scalability.

Main Results:

• SGCAST demonstrated improved accuracy in spatial domain identification on benchmarking datasets.
• Validated SGCAST's performance across various scales and multiple ST platforms.
• Showcased SGCAST's superior capacity for analyzing large-scale, high-resolution ST data.

Conclusions:

• SGCAST offers an efficient and scalable solution for spatial domain identification in ST data.
• The framework enhances the analysis of gene expression within the tissue microenvironment.
• SGCAST represents a significant advancement in computational tools for spatial transcriptomics.