SpatialDG: a novel spatial domain identification method for spatially resolved transcriptomics data based on...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatially resolved transcriptomic (ST) technologies provide gene expression data within tissue architecture.
• Identifying discrete biological domains is crucial for understanding tissue heterogeneity, development, and disease.
• ST data's noise, high dimensionality, and sparsity pose challenges for unsupervised domain delineation.

Purpose of the Study:

• To develop an unsupervised framework for delineating biologically meaningful domains from ST data.
• To address the challenges of noise, high dimensionality, and sparsity in ST data.
• To provide a robust and generalizable tool for analyzing ST datasets.

Main Methods:

• Introduced SpatialDG, a dual-graph self-supervised contrastive learning framework for ST data.
• Utilized graph neural networks and self-supervised contrastive learning to learn spot representations.
• Integrated gene expression similarity and spatial adjacency graphs via a dual-view contrastive architecture.
• Employed a zero-inflated negative binomial reconstruction loss for count-based and sparse data.

Main Results:

• SpatialDG demonstrated significant performance gains over existing state-of-the-art algorithms.
• The framework achieved robust generalization across diverse ST datasets, including healthy and cancer tissues.
• SpatialDG successfully identified biologically meaningful domains by integrating spatial and genetic signals.

Conclusions:

• SpatialDG efficiently unravels biologically meaningful domains from ST data.
• The framework provides a powerful and generalizable tool for mining tissue architecture.
• SpatialDG advances the analysis of complex biological systems using ST technologies.