Comparative analysis of dimension reduction methods for cytometry by time-of-flight data

Area of Science:

• Computational biology
• Bioinformatics
• Single-cell analysis

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis methods are advanced, but mass cytometry (CyTOF) data analysis tools lag significantly.
• CyTOF data possess unique characteristics necessitating specialized computational approaches.
• Dimension reduction (DR) is a crucial step in analyzing single-cell data.

Purpose of the Study:

• To benchmark the performance of 21 dimension reduction (DR) methods for mass cytometry (CyTOF) data.
• To identify optimal DR methods for CyTOF data analysis based on real and synthetic datasets.
• To provide guidance on selecting appropriate DR methods based on data structure and analytical goals.

Main Methods:

• Benchmarking of 21 DR methods applied to 110 real and 425 synthetic CyTOF samples.
• Evaluation of DR method performance based on structure preservation and downstream analysis utility.
• Comparative analysis of methods including SAUCIE, SQuaD-MDS, scvis, UMAP, and t-SNE.

Main Results:

• SAUCIE, SQuaD-MDS, and scvis demonstrated superior overall performance in CyTOF data analysis.
• SAUCIE and scvis offer a balanced performance profile.
• SQuaD-MDS excels in preserving data structure, while UMAP shows strong downstream analysis performance. t-SNE and its hybrid variants provide the best local structure preservation.

Conclusions:

• The choice of DR method for CyTOF data analysis should be tailored to specific data characteristics and research objectives.
• A high degree of complementarity exists among different DR tools, suggesting integrated approaches may be beneficial.
• Further development of computational methods specific to CyTOF data is warranted to match the pace of experimental advancements.