Quantitative Comparison of Conventional and t-SNE-guided Gating Analyses

Area of Science:

• Immunology
• Computational Biology
• Data Science

Background:

• High-parameter single-cell data analysis is crucial for understanding immune cell heterogeneity.
• t-Distributed Stochastic Neighbor Embedding (t-SNE) is a popular dimensionality reduction technique for data visualization.
• The comparison between t-SNE and manual gating for immune cell quantification remains underexplored.

Purpose of the Study:

• To compare the efficacy of t-SNE analysis versus conventional manual hand-gating for stratifying and quantifying immune cell populations.
• To assess the consistency and discrepancies between t-SNE-guided and manual gating approaches in immune cell subset identification.

Main Methods:

• Application of a 38-parameter mass cytometry panel to human blood samples.
• Comparison of immune cell subset frequencies using both conventional bivariate gating and t-SNE-guided manual gating.
• Analysis of 28 distinct immune cell subsets.

Main Results:

• t-SNE analysis successfully stratified general cellular lineages and most sub-lineages with high correlation to manual gating.
• Discrepancies were observed for specific immune cell subsets, where manual gating of continuous variables resulted in intermingled populations in t-SNE space.
• High correlation was found between conventional and t-SNE-guided cell frequency calculations for most subsets.

Conclusions:

• t-SNE demonstrates consistency with conventional hand-gating for stratifying major immune cell lineages.
• Certain immune cell subsets, particularly those defined by continuous variables in manual gating, may not be fully resolved in t-SNE space, leading to identification and quantification differences.
• The study highlights both the strengths and limitations of t-SNE in immune cell subset analysis, suggesting careful consideration for specific subset delineation.