.png){kind=icon}
.png){kind=icon}
Return
A multi-View graph attention network for spatial domain identification in spatial transcriptomics
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.134145.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
•
Propose a novel framework named ST-GATE for spatial domain identification.
•
Leverage a three-level CL strategy to enhance embedding discriminability.
•
A graph-adaptive spatial attention method is designed to improve feature learning.
•
The effectiveness of ST-GATE is validated on multiple real-world datasets.
Abstract
Spatial transcriptomics (ST) is an emerging technology that profiles gene expression in tissues while preserving spatial location information. Linking cellular gene expression to spatial distribution is pivotal for accurate spatial domain identification, deepening the understanding of tissue microenvironments and biological processes. However, effectively integrating gene expression data with spatial information to identify spatial domains remains challenging. To address these challenges, we propose ST-GATE, an innovative deep learning framework that synergistically combines multi-view graph convolutional networks with graph-adaptive spatial attention mechanisms and contrastive learning for spatial domain recognition. First, ST-GATE adapts to ST data of varying resolutions by constructing two complementary neighborhood graphs that incorporate local features from k-nearest neighbors and global features from r-radius. Subsequently, a graph convolutional network is employed to embed gene expression information into spatially informed latent representations. To refine the learned embeddings, ST-GATE incorporates a hierarchical contrastive learning framework that collaboratively captures spatial features from local, global, and contextual levels. Furthermore, ST-GATE leverages graph-adaptive spatial attention to enhance the GCN-derived representations, thereby learning the significance of different features. Ultimately, we employ an attention network to fuse multi-view outputs to reconstruct the final gene expression representation. We compare ST-GATE with other state-of-the-art spatial domain identification methods on four real ST datasets from different tissues. Extensive experimental results demonstrate that ST-GATE achieves superior prediction performance.
Keywords:
Spatial transcriptomics
Graph-adaptive spatial attention mechanism
Spatial domain identification
Multi-view graph convolutional networks
Hierarchical contrastive learning
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
View more organizations