stUAI: Uncertainty-Aware Clustering of Spatially Resolved Transcriptomics Data

Wei Ju, Lezhi Chen, Siyu Yi, Wei Zhang, Yifan Wang, Yanran Zhu, Ziyue Qiao, Jiancheng Lv

Keywords: Data Clustering, Spatial Transcriptomics, Uncertainty Learning, Graph Neural Networks

Abstract: Spatially resolved transcriptomics (SRT) technologies enable high-resolution investigation of gene expression patterns across tissue sections, providing unprecedented insights into the molecular architecture of tissues. However, the inherent sparsity and over-dispersion of gene expression across spots, together with view-specific heterogeneity, jointly complicate modeling and present formidable challenges to reliable spatially informed downstream analyses. To address these issues, we propose stUAI, an Uncertainty-AwareIntegration framework for spatially transcriptomics data. stUAI first learns spatial- and expression-view embeddings by two separate graph-based encoders to capture multi-scale information. Then stUAI exploits the distributional representation of spots to quantify the view-specific uncertainty caused by technology limitations or noise in SRT, which is further leveraged for intra-view contrastive learning, cross-view information alignment, and representation integration. Furthermore, stUAI incorporates a zero-inflated negative binomial decoder to handle expression sparsity and imposes spatial structural constraints to preserve spatial continuity. Extensive experimental results on multiple benchmark datasets validate the effectiveness of our proposed stUAI in spatial clustering and several downstream applications.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 3635