DRjCC

DRjCC integrates joint dimension reduction and cell clustering to improve identification of cell types from single-cell RNA-sequencing (scRNA-seq) data by addressing noise, high-dimensionality, and linear inseparability.

Key Features:

• Joint learning framework: Simultaneously performs dimension reduction and clustering so that feature selection during reduction is informed by clustering structure.
• Dimension reduction (projected matrix decomposition): Reduces scRNA-seq data complexity via projected matrix decomposition while preserving essential features.
• Cell clustering (non-negative matrix factorization): Identifies and categorizes cell types and subtypes using non-negative matrix factorization.
• Optimization strategy: Formulates the joint learning process as a constrained optimization problem and derives specific optimization rules to improve reduction and clustering accuracy.
• Performance validation: Evaluated on eleven scRNA-seq datasets (49–68,579 cells; 3–14 cell types) and reported an average improvement of 17.44% compared with thirteen state-of-the-art methods.
• Robustness and efficiency: Demonstrated robustness and efficiency across datasets from various tissues.

Scientific Applications:

• Cell type discovery: Discovery of cell types and subtypes from scRNA-seq data.
• Cellular heterogeneity analysis: Analysis and characterization of cellular heterogeneity within tissues.
• Novel population identification: Identification of novel cell populations in complex samples.
• Tissue structure elucidation: Elucidation of complex tissue structures through improved cell-type resolution.

Methodology:

Dimension reduction via projected matrix decomposition, clustering via non-negative matrix factorization, and formulation of the joint learning task as a constrained optimization problem with derived optimization rules.