scSiameseClu: A Siamese Clustering Framework for Interpreting single-cell RNA Sequencing Data

TL;DR

scSiameseClu tackles the challenges of noisy, sparse, and high-dimensional scRNA-seq data by introducing a three-module Siamese clustering framework. The Dual Augmentation Module creates robust, informative views at both gene and cell-graph levels; the Siamese Fusion Module performs cross-view refinement and adaptive feature fusion to avoid over-smoothing; and the Optimal Transport Clustering aligns cluster assignments with predefined proportions using Sinkhorn distance. Together, these components yield discriminative embeddings that improve clustering accuracy and enable reliable downstream analyses such as cell-type annotation and marker gene identification. Evaluations on seven real datasets show consistent improvements over state-of-the-art methods in clustering metrics and biological interpretability, underscoring the method’s practical impact for single-cell transcriptomics.

Abstract

Single-cell RNA sequencing (scRNA-seq) reveals cell heterogeneity, with cell clustering playing a key role in identifying cell types and marker genes. Recent advances, especially graph neural networks (GNNs)-based methods, have significantly improved clustering performance. However, the analysis of scRNA-seq data remains challenging due to noise, sparsity, and high dimensionality. Compounding these challenges, GNNs often suffer from over-smoothing, limiting their ability to capture complex biological information. In response, we propose scSiameseClu, a novel Siamese Clustering framework for interpreting single-cell RNA-seq data, comprising of 3 key steps: (1) Dual Augmentation Module, which applies biologically informed perturbations to the gene expression matrix and cell graph relationships to enhance representation robustness; (2) Siamese Fusion Module, which combines cross-correlation refinement and adaptive information fusion to capture complex cellular relationships while mitigating over-smoothing; and (3) Optimal Transport Clustering, which utilizes Sinkhorn distance to efficiently align cluster assignments with predefined proportions while maintaining balance. Comprehensive evaluations on seven real-world datasets demonstrate that scSiameseClu outperforms state-of-the-art methods in single-cell clustering, cell type annotation, and cell type classification, providing a powerful tool for scRNA-seq data interpretation.

Figures (13)

• Figure 1: Similarity distributions of cell embeddings learned by scNAME and scGNN on dataset Human liver cells.
• Figure 2: Overview architecture of scSiameseClu. It contains three components: (i) a data-augmented module, (ii) a Siamese fusion module, and (iii) an optimal transport clustering strategy for self-supervision learning.
• Figure 3: Visualization of scSiameseClu and four typical baselines on human liver cells in 2D t-SNE projection. Each point represents a cell, while each color represents a predicted cell type.
• Figure 4: Distribution plot and heat map of cell similarities in latent space learned by scSiameseClu on the Human liver cells dataset.
• Figure 5: Cell type annotation: overlap of top 50 DEGs in clusters versus gold standard cell types (similarity = overlapping DEGs/50).