Cluster stability in the analysis of mass cytometry data.

Area of Science:

• Single-cell analysis
• Computational biology
• Immunology

Background:

• Manual gating of cytometry data is traditional but limited with high-dimensional mass cytometry.
• Automated clustering offers a solution for analyzing complex, high-dimensional single-cell data.
• Reproducibility of automated clustering results remains a challenge, necessitating robust interpretation frameworks.

Purpose of the Study:

• To evaluate the importance and application of cluster stability metrics in mass cytometry data analysis.
• To assess the stability of clusters generated by popular algorithms (SPADE, FLOCK, PhenoGraph).
• To propose cluster stability as a key checkpoint for rigorous interpretation of cytometry clustering results.

Main Methods:

• Applied three clustering algorithms (SPADE, FLOCK, PhenoGraph) to four mass cytometry datasets.
• Evaluated cluster stability based on consistent re-occurrence within and between algorithms.
• Compared automated clustering results with manual gating to assess biological relevance.
• Investigated relationships between cluster stability, compactness, and isolation.

Main Results:

• Clustering algorithms produced varying degrees of statistical stability, with many unstable clusters identified.
• Cluster stability is linked to compactness and isolation but cannot be reliably predicted by these properties alone.
• Cluster stability information aids in distinguishing biologically relevant clusters from spurious ones.

Conclusions:

• Cluster stability evaluation is essential for reproducible and reliable interpretation of mass cytometry data.
• The study advocates for incorporating cluster stability as a standard analytical checkpoint.
• This work contributes to a more systematic framework for assessing cytometry clustering outcomes.