Leveraging Spot-Gene Heterogeneous Graphs for Unified Spatially Resolved Transcriptomics Domain Detection on...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Existing spatially resolved transcriptomics (SRT) domain detection methods struggle with spot proximity reliance, multi-slice alignment, and batch effects.
• Accurate spatial domain identification is crucial for understanding tissue architecture and function.

Purpose of the Study:

• To develop a unified and accurate method for spatial domain identification in both single-slice and multi-slice SRT datasets.
• To overcome the limitations of current SRT domain detection techniques, including batch effect mitigation and alignment requirements.

Main Methods:

• Proposed spatially resolved transcriptomics heterogeneous graph contrastive learning (stHGCL), integrating a spot-gene heterogeneous graph and a dual-stage encoder (LightGCN and GCN).
• Employed neighborhood-driven contrastive learning to refine spot embeddings, enhance intra-cluster compactness, and mitigate batch effects.
• Utilized heterogeneous graphs to capture high-order structural information via spot-gene connections.

Main Results:

• stHGCL demonstrated superior performance on seven benchmark SRT datasets across multiple platforms (10x Visium, BaristaSeq, STARmapSeq, Slide-seq, Stereo-seq).
• Outperformed nine single-slice and eight multi-slice state-of-the-art methods, achieving top Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI) scores.
• Successfully delineated complex spatial domains and enabled cross-slice detection for unaligned multi-slice datasets.

Conclusions:

• stHGCL effectively captures high-order structural and spatial information while mitigating batch effects.
• Provides a robust and scalable solution for unified spatial domain detection in SRT.
• Facilitates deeper insights into spatial domains across diverse SRT experimental paradigms.