A Novel Low-Dimensional Sparse and Low-Rank Representation Method for Single-Cell RNA Sequencing Data Clustering

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular heterogeneity, crucial for biology and medicine.
• Accurate cell clustering is vital for scRNA-seq data analysis but challenged by high dimensionality and sparsity.
• Existing methods struggle to precisely cluster cells due to data complexities.

Purpose of the Study:

• To develop a novel, end-to-end computational model for accurate cell clustering of scRNA-seq data.
• To address the challenges of high dimensionality and sparsity in scRNA-seq datasets.
• To improve the precision of cell type identification and biological discovery from scRNA-seq.

Main Methods:

• Proposes LDSLRR (Low-Dimensional Sparse and Low-Rank Representation), an integrated model.
• Employs dimensionality reduction via projection of the gene expression matrix.
• Learns a self-representation matrix using low-rank and sparsity constraints.
• Performs clustering via graph-regularized non-negative matrix factorization (NMF).
• Simultaneously optimizes dimensionality reduction, self-representation, and clustering modules.

Main Results:

• LDSLRR demonstrates superior performance compared to state-of-the-art clustering methods.
• Experiments on multiple scRNA-seq datasets validate the effectiveness of the proposed approach.
• The integrated end-to-end model enhances clustering accuracy.

Conclusions:

• LDSLRR offers a robust and accurate solution for cell clustering in scRNA-seq data.
• The method effectively overcomes inherent data challenges, improving biological interpretation.
• This approach advances the analysis of cellular heterogeneity for biological research.