Identifying cell types by lasso-constraint regularized Gaussian graphical model based on weighted distance penalty.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Existing cell identification methods often lack accuracy or require specific equipment.
• Unsupervised methods typically need the number of cell types specified beforehand, limiting their use.

Purpose of the Study:

• To develop a novel, accurate, and widely applicable algorithm for cell type identification from scRNA-seq data.
• To overcome the limitations of existing unsupervised clustering methods, particularly the need for pre-specifying cell numbers.
• To provide a robust foundation for downstream biological analyses.

Main Methods:

• Proposed the WLGG algorithm framework for scRNA-seq data analysis.
• Incorporated a weighted distance penalty with a Gaussian kernel to capture nonlinear data information.
• Applied a Lasso constraint on a regularized Gaussian graphical model for linear data characteristics.
• Utilized the Eigengap strategy for automatic cell type number prediction and spectral clustering for label assignment.

Main Results:

• WLGG demonstrated superior clustering accuracy compared to 16 alternative methods across 14 test datasets.
• Downstream analyses, including marker gene identification and pseudotime inference, confirmed WLGG's reliability.
• The algorithm provided valuable insights into dynamic biological processes and regulatory mechanisms.

Conclusions:

• The WLGG algorithm offers a significant advancement in accurate and automated cell type identification from scRNA-seq data.
• Its ability to predict cell numbers and handle nonlinear data enhances its applicability in biological research.
• WLGG facilitates deeper understanding of cellular heterogeneity and complex biological systems.