Mixture models for single-cell assays with applications to vaccine studies

Area of Science:

• Immunology
• Computational Biology
• Biostatistics

Background:

• Single-cell assays are crucial for dissecting complex biological systems by measuring distinct cell subsets.
• Characterizing small cell subsets requires accurate single-cell measurements of multiple genes and proteins.
• Identifying differentially expressed biomarkers in these subsets is vital for understanding biological changes.

Purpose of the Study:

• To develop a robust statistical framework for testing differential biomarker expression in single-cell assays.
• To provide a method that accounts for subject-specific variations, essential for studies like vaccine response assessment.
• To offer improved sensitivity and specificity compared to existing methods for biomarker analysis.

Main Methods:

• A Bayesian hierarchical framework utilizing a beta-binomial mixture model for differential biomarker expression testing.
• Two parameter estimation approaches: empirical-Bayes (Expectation-Maximization) and fully Bayesian (Markov chain Monte Carlo).
• Extension to multivariate differential expression using a Dirichlet-multinomial model.

Main Results:

• The proposed Bayesian method demonstrates higher sensitivity and specificity in identifying differential biomarker expression compared to classical approaches like Fisher's exact test.
• Simulations confirm the framework's robustness to model misspecification.
• The method effectively analyzes single-cell gene and protein expression data, including multivariate analyses.

Conclusions:

• The Bayesian hierarchical framework offers a powerful and accurate approach for analyzing single-cell assay data.
• This method improves the identification of biomarkers indicative of biological changes, particularly in subject-specific studies.
• The framework is versatile and can be extended for complex multivariate differential expression analyses.