Determining clinically relevant features in cytometry data using persistent homology.

Area of Science:

• Immunology
• Computational Biology
• Data Science

Background:

• Cytometry generates high-dimensional point cloud data challenging manual interpretation.
• Standard Boolean gating misses subtle topological features, especially with batch effects or donor variations.
• Existing methods struggle to capture complex data structures in clinical cytometry datasets.

Approach:

• Applied persistent homology, a topological data analysis method, to cytometry data.
• Utilized a decision-tree classifier and kernel-density estimator to identify key proteins and sample data points.
• Computed persistence diagrams and Wasserstein distances to quantify structural differences between patient groups.

Key Points:

• Persistent homology effectively distinguishes data sources, like healthy donors versus COVID-19 patients.
• Identified significant structural differences in T-bet, Eomes, and Ki-67 expression in COVID-19 patients.
• Found downregulation of T-bet and Eomes in non-naïve CD8+ T cells of COVID-19 patients.

Conclusions:

• Topological data analysis reveals novel insights into cytometry data, overcoming limitations of standard gating strategies.
• The findings suggest a reduced prevalence of canonical effector CD8+ T cells in COVID-19 patients.
• This approach is broadly applicable to cytometry datasets for uncovering hidden patterns and batch effect-resilient analysis.