Hypergraph-Enhanced Self-Supervised Learning with Module-Aware Structured Masking for Single-Cell Transcriptomic Clustering.

Accurate clustering of single-cell RNA sequencing (scRNA-seq) data is essential for cell type annotation and cellular heterogeneity inference, but existing methods are still limited by local gene co-expression redundancy and pairwise graph modeling, which hinder the capture of cross-module biological semantics and higher-order topological cellular relationships. To address these challenges, we propose a novel method called scHyperMod. It employs a module-aware structured masking strategy that suppresses shortcut learning driven by local co-expression redundancy and encourages the model to learn more discriminative representations from cross-module contexts. Furthermore, hypergraph modeling based on shared nearest neighbors is introduced to enable information propagation across locally connected cells, thus providing a more expressive description of higher-order topology. Extensive evaluations on 12 public benchmark datasets show that scHyperMod achieves leading overall clustering performance across varying data scales, sparsity levels, and cellular heterogeneity. Application to a glioblastoma dataset further demonstrates its ability to recover major cell populations and distinguish malignant cell states, including AC-like, OPC-like, and Cycling, while preserving cell identity-related biological differences in marker gene expression and functional enrichment.