Accuracy, robustness and scalability of dimensionality reduction methods for single-cell RNA-seq analysis

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Dimensionality reduction is crucial for single-cell RNA sequencing (scRNA-seq) data analysis, aiding noise removal and downstream tasks like cell clustering and lineage reconstruction.
• Despite numerous methods, few comprehensive comparisons exist to evaluate their effectiveness in scRNA-seq.

Purpose of the Study:

• To address the knowledge gap by comparatively evaluating common dimensionality reduction methods for scRNA-seq.
• To provide guidelines for selecting appropriate dimensionality reduction techniques.

Main Methods:

• Compared 18 dimensionality reduction methods across 30 diverse scRNA-seq datasets.
• Evaluated neighborhood preservation, cell clustering accuracy and robustness, and lineage reconstruction.
• Assessed computational scalability and cost of each method.

Main Results:

• Performance varied significantly across methods and datasets.
• Specific methods excelled in neighborhood preservation, while others were better for clustering and lineage reconstruction.
• Computational cost was a key differentiating factor.

Conclusions:

• Comprehensive evaluation provides crucial guidelines for choosing optimal dimensionality reduction methods for scRNA-seq data.
• Analysis scripts are available to ensure reproducibility and facilitate further research.