Joint learning dimension reduction and clustering of single-cell RNA-sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cell type discovery through unsupervised clustering.
• Current methods often perform dimension reduction separately from clustering, limiting performance due to data noise and complexity.
• This separation fails to capture intricate patterns essential for accurate cell type identification.

Purpose of the Study:

• To introduce DRjCC, a novel algorithm for scRNA-seq data analysis.
• To improve cell type discovery by jointly learning dimension reduction and cell clustering.
• To provide a flexible and accurate computational strategy for scRNA-seq data.

Main Methods:

• Developed DRjCC, a method that integrates dimension reduction via projected matrix decomposition and cell clustering using non-negative matrix factorization.
• Formulated the joint learning process as a constrained optimization problem with derived optimization rules.
• Utilized feature selection in dimension reduction guided by cell clustering for enhanced performance.

Main Results:

• DRjCC demonstrated superior performance across eleven diverse scRNA-seq datasets, outperforming 13 state-of-the-art methods.
• Achieved an average improvement of 17.44% in cell type clustering accuracy.
• Showcased efficiency and robustness across datasets with varying cell numbers and tissue origins.

Conclusions:

• DRjCC offers a significant advancement in analyzing scRNA-seq data for cell type discovery.
• The joint learning approach effectively addresses the limitations of sequential dimension reduction and clustering.
• The algorithm provides a robust and efficient strategy for uncovering cellular heterogeneity.