Hierarchical modeling for rare event detection and cell subset alignment across flow cytometry samples.

Area of Science:

• Immunology
• Computational Biology
• Biostatistics

Background:

• Flow cytometry is crucial for analyzing antigen-specific lymphocytes in vaccine and biomarker research.
• Manual analysis of flow cytometry data is subjective and difficult to reproduce.
• Automated methods are needed to detect rare cell subsets and align data across samples.

Purpose of the Study:

• To develop an automated, objective method for identifying rare cell subsets in flow cytometry data.
• To improve the detection sensitivity and reproducibility of antigen-specific lymphocyte enumeration.
• To enable robust comparative analysis across multiple flow cytometry samples.

Main Methods:

• Developed hierarchical extensions to the Dirichlet Process Gaussian Mixture Model (DPGMM), termed HDPGMM.
• Utilized HDPGMM for automated cell subset identification and alignment across samples.
• Validated HDPGMM using clinically relevant peripheral blood mononuclear cell (PBMC) samples with known frequencies of antigen-specific T cells.

Main Results:

• HDPGMM provides an aligned data model capturing commonalities and variations across samples.
• The method increases sensitivity for detecting extremely low-frequency events by sharing information across samples.
• Validated accuracy and reproducibility of HDPGMM for quantifying antigen-specific T cells.

Conclusions:

• Hierarchical modeling offers a robust probabilistic approach for flow cytometry data analysis.
• HDPGMM ensures consistent cell subset labeling and enhances rare event detection.
• The developed open-source software leverages multi-processor and GPU acceleration for demanding computations.