.png){kind=icon}
.png){kind=icon}
Return
Integrating modularity maximization and contrastive learning for identifying spatial domain from spatial transcriptomics
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2025.129426.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose a novel framework named STMACL, which combines the advantages of generative and contrastive to effectively fuse gene expression profiles and spatial context for spatial domain identification. • We introduce an MMC module, which leverages modularity maximization as a pretext task of contrastive learning to mine the reliable embeddings and performs two-stage random walk strategies to capture high-order proximity. • We design an ICR strategy to improve feature discriminability. Furthermore, an REC module is constructed to simultaneously reconstruct the original expression matrix and the spatial topology. This strategy ensures that the model learns latent representations that better preserve the original input information. • We conduct comprehensive comparisons with 8 state-of-the-art methods on 7 public ST datasets. Experimental results demonstrate that STMACL outperforms baselines and achieves superior performance.
Keywords:
STMACL
contrastive learning
gene expression profiles
spatial domain identification
modularity maximization
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W