CellMentor: cell-type aware dimensionality reduction for single-cell RNA-sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional gene expression data.
• Dimensionality reduction is crucial for analyzing scRNA-seq data, enabling clustering and cell type identification.
• Existing methods struggle to effectively remove technical noise while preserving biological signals.

Purpose of the Study:

• To develop a novel, fully supervised dimensionality reduction method for scRNA-seq data.
• To create a tool that optimizes low-dimensional embeddings for accurate cell type identification.
• To improve the balance between noise removal and biological signal preservation in scRNA-seq analysis.

Main Methods:

• Introduced CellMentor, a supervised dimensionality reduction technique.
• Utilized non-negative matrix factorization (NMF) integrated with cell type labels.
• Incorporated an optimization objective that minimizes within-group variation and maximizes between-group distinctions.

Main Results:

• CellMentor demonstrated superior cell type separation on diverse datasets.
• The method showed robust performance in batch correction across experiments.
• CellMentor effectively identified rare cell populations, outperforming existing approaches.
• Generated low-dimensional embeddings optimized for cell type identification.

Conclusions:

• CellMentor offers a powerful, supervised approach for scRNA-seq dimensionality reduction.
• The method enhances the accuracy of cell type identification and rare cell detection.
• CellMentor provides a valuable tool for integrative single-cell analyses and biological discovery.