A deep matrix factorization based approach for single-cell RNA-seq data clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell sequencing technologies enable detailed analysis of cellular heterogeneity.
• Accurate cell clustering is crucial for understanding this heterogeneity but remains challenging due to data noise, high dimensionality, and sparsity.
• Existing methods struggle to effectively address these inherent data complexities.

Purpose of the Study:

• To introduce DeepCI, a novel computational approach for robust cell clustering in single-cell RNA sequencing (scRNA-seq) data.
• To simultaneously learn low-dimensional cell representations and perform accurate clustering.
• To evaluate DeepCI's performance against established single-cell analysis methods.

Main Methods:

• DeepCI utilizes two autoencoders to generate cell and gene embeddings, facilitating joint learning of low-dimensional representations and clustering.
• It reconstructs the gene expression matrix via matrix multiplication of cell and gene embeddings.
• Performance was validated on multiple real-world scRNA-seq datasets for clustering, visualization, and gene expression imputation.

Main Results:

• DeepCI demonstrated superior performance in clustering and visualization compared to several popular single-cell analysis techniques.
• The method effectively imputes gene expression, significantly improving cell type classification accuracy when using known marker genes.
• Experimental results confirm DeepCI's capability to handle noisy and high-dimensional scRNA-seq data effectively.

Conclusions:

• DeepCI offers a powerful and accurate solution for cell clustering and data imputation in scRNA-seq analysis.
• The approach effectively addresses the inherent challenges of single-cell data, leading to improved biological insights.
• Imputation of marker gene expression by DeepCI is a valuable strategy for enhancing cell type identification.