Structure-Guided Integrative Soft Deep Clustering Analysis of scRNA-Seq and scATAC-Seq Data

Xingzuo Jiang¹Chenyuan Wang¹Jiaxi Yao¹Chengyuan Wang^1,2*

• ¹Department of Urology, The First Hospital of China Medical University, Shenyang, China
• ²Department of Epidemiology, China Medical University, Shenyang, China

Single-cell sequencing technologies, such as scRNA-seq and scATAC-seq, have enabled significant advancements in understanding cellular diversity. However, current methods force the cell to be assigned to hard clustering labels, which is inappropriate within the single-cell development scenario. We propose the Structure-Guided Soft Deep Clustering (sgSDC) framework, which enhances clustering performance by utilizing inter-sample and inter-modality information. Specifically, it introduces a contrastive learning module that aligns modality-specific representations with a consensus representation, reducing modality-specific noise. Furthermore, the model incorporates a soft clustering mechanism, allowing cells to associate probabilistically with multiple clusters to capture their dynamic developmental states. Extensive evaluations on four benchmark datasets demonstrate that sgSDC outperforms eight state-of-the-art methods in clustering accuracy, normalized mutual information, and adjusted Rand index, offering a robust tool for exploring cellular heterogeneity and advancing research in tumor microenvironments.