A Quantitative Framework for Evaluating Single-Cell Data Structure Preservation by Dimensionality Reduction...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional single-cell RNA sequencing (scRNA-seq) data pose interpretation and visualization challenges.
• Dimensionality reduction techniques are vital for analyzing scRNA-seq data, enabling downstream tasks like clustering and trajectory reconstruction.
• A quantitative evaluation of these dimensionality reduction methods is currently lacking.

Purpose of the Study:

• To establish an unbiased framework for evaluating dimensionality reduction methods on scRNA-seq data.
• To define metrics for assessing the preservation of global and local data structures.
• To quantitatively compare the performance of 11 common dimensionality reduction methods.

Main Methods:

• Developed a framework with metrics for global and local structure preservation.
• Utilized discrete and continuous real-world and synthetic scRNA-seq datasets.
• Analyzed the impact of input cell distribution and method parameters on structure preservation.

Main Results:

• Input cell distribution and method parameters significantly influence structure preservation.
• Performance varied across 11 common dimensionality reduction methods.
• The framework provides quantitative insights into method-specific performance.

Conclusions:

• The proposed framework enables objective evaluation of dimensionality reduction techniques for scRNA-seq data.
• Understanding the impact of data distribution and parameters is key to selecting appropriate methods.
• This work facilitates more reliable biological interpretation of high-dimensional single-cell data.