sccomp

sccomp performs differential composition and variability analyses for cellular omics data by jointly modeling data count distribution, compositionality, group-specific variability, and proportion mean-variability association with outlier awareness.

Key Features:

• Bayesian Modeling: Provides robust Bayesian modeling with outlier detection, random effects, and advanced statistical methods for cell type proportion analysis.
• Probabilistic Outlier Identification: Identifies outliers probabilistically as part of the modeling framework.
• Mixed-Effect Modeling: Incorporates mixed-effect modeling to account for hierarchical or correlated data structures.
• Differential Variability Testing: Enables dedicated statistical testing for differential variability in composition, which was previously lacking for cellular omics data.
• Realistic Data Simulation: Offers a framework for realistic data simulation based on the fitted model parameters.
• Cross-Study Knowledge Transfer: Supports transfer of knowledge across different studies through the modeling framework.

Scientific Applications:

• Cancer Research: Identifies differential constraints and composition in the tumor microenvironment, as demonstrated in primary breast cancer analysis.
• Immunology: Enables characterization and comparison of immune cell composition between conditions to identify markers of disease progression.
• Developmental Biology: Supports analysis of cell type proportion changes during development through differential composition analysis.
• Single-Cell Genomics: Analyzes single-cell RNA sequencing data for differential composition and variability.
• Proteomics: Applies to single-cell proteomics data for compositional analysis.
• Microbiomics: Analyzes microbiome data for differential composition and variability of microbial communities.
• Pathogen Infection Studies: Identifies markers of pathogen infection through comparative composition analysis.

Methodology:

sccomp jointly models data count distribution, compositionality, group-specific variability, and proportion mean-variability association, being aware of outliers. The method uses robust Bayesian modeling with outlier detection and random effects, and does not use the Dirichlet-multinomial distribution due to the inadequacy of that distribution given ubiquitous mean-variability association across technologies.