基于图正则化非负矩阵分解的scATAC-seq数据分析

Analyzing Single-Cell ATAC-seq Data with Graph Regularized Non-Negative Matrix Factorization

摘要: 针对scATAC-seq数据的稀疏性与噪声问题，本研究提出一种融合图正则化与核范数约束的非负矩阵分解模型。该模型通过矩阵分解恢复全局信息，并利用细胞相似性图保持局部流形结构，从而在低维空间中实现生物一致性表示。实验表明，本方法能有效恢复缺失值，并在聚类与可视化中揭示更清晰的细胞亚群，提升下游分析性能。

Abstract: To address the challenges of sparsity and technical noise in single-cell ATAC sequencing data, this study proposes a non-negative matrix factorization method integrated with graph regularization and nuclear norm constraint. This approach restores global information through low-rank decomposition and preserves the local manifold structure by leveraging a cell similarity graph, thereby achieving biologically consistent representations in the low-dimensional space. Experiments demonstrate that our method effectively recovers missing values, reveals clearer cell subpopulations in clustering and visualization tasks, and enhances the performance of downstream scATAC-seq data analysis.