Path-MGCN: a pathway activity based multi-view graph convolutional network for determining spatial domains with attention mechanism
Qirui Zhou, Chaowen Li, Chao Chen, Mingyue Li, Jiabei Liu, Weijun Sun, Zongmeng Zhang, Songqing Gu, Yishan Cai, Yonghui Huang, Hongtao Liu, Chao Yang, Xin Chen · Research Square · 2024
Abstract Gene functional relationships are always ignored in spatial-domain recognition based on spatial transcriptomics (ST). We develop Path-MGCN, a multi-view graph convolutional network (MGCN) with attention mechanism that embeds pathway information. We generate a pathway activity profile with spot-specific pathway enrichment. Unique and shared embeddings from pathway and spatial graphs are extracted by a MGCN encoder, dynamically optimized by attention mechanism, followed by a decoder to retain the original pathway information. Path-MGCN outperforms state-of-the-art spatial clustering methods. Moreover, Path-MGCN could identify spatial domain-specific pathways for further mechanism study in the context of microenvironment, enabling the precision medicine of complex diseases.