A Bayesian feature allocation model for identifying cell subpopulations using CyTOF data
Arthur Lui1, Juhee Lee2, Peter F Thall3
1Department of Statistics, Baskin School of Engineering, University of California Santa Cruz, 1156 High Street, Santa Cruz, CA, 95064, USA.
Summary
A new Bayesian model identifies cell subpopulations from cytometry data across multiple samples. This approach improves the discovery of novel cell subsets, particularly for natural killer (NK) cells, aiding cancer immunotherapy research.
Area of Science:
- Computational biology
- Immunology
- Statistical modeling
Background:
- Cytometry by time of flight (CyTOF) generates high-dimensional data for cell analysis.
- Identifying distinct cell subpopulations is crucial for understanding cellular functions and disease states.
- Existing methods often analyze samples independently, potentially missing cross-sample patterns.
Purpose of the Study:
- To develop a Bayesian feature allocation model (FAM) for robust cell subpopulation identification.
- To enable simultaneous analysis of multiple CyTOF datasets.
- To account for technical artifacts like missing data in mass cytometry.
Main Methods:
- Bayesian feature allocation model (FAM) using a finite Indian buffet process.
- Joint analysis of multiple CyTOF samples.
- Incorporation of a static missingship mechanism for non-ignorable missing data.
Main Results:
- The FAM effectively clusters cells into subpopulations based on marker expression patterns.
- Simultaneous multi-sample analysis identified cell subpopulations missed by conventional methods.
- Application to natural killer (NK) cell data revealed potential novel NK cell subsets.
Conclusions:
- The FAM provides a powerful framework for dissecting cellular heterogeneity in multi-sample CyTOF data.
- Identified novel NK cell subsets may offer insights into NK cell biology and cancer immunotherapy.
- This statistical approach can advance the development of improved NK cell-based therapies.
