tascCODA: Bayesian Tree-Aggregated Analysis of Compositional Amplicon and Single-Cell Data

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Modeling

Background:

• High-throughput sequencing generates count data (e.g., microbiome, single-cell RNA-seq) requiring accurate statistical modeling.
• Compositional data with hierarchical structures (taxonomic, lineage trees) present unique analytical challenges.

Purpose of the Study:

• Introduce tascCODA, a Bayesian model for compositional count data analysis.
• Integrate hierarchical information and covariates for generative modeling.
• Enable data-driven, parsimonious determination of covariate effects across population hierarchies.

Main Methods:

• Developed a Bayesian model (tascCODA) for tree-aggregated compositional count data.
• Integrated latent tree structure parameters with spike-and-slab Lasso penalization.
• Validated performance using synthetic benchmark scenarios and real-world datasets.

Main Results:

• tascCODA accurately models compositional count data with hierarchical structures.
• Identified aggregated cell type and taxon compositional changes in ulcerative colitis and irritable bowel syndrome data.
• Achieved more predictive and parsimonious results compared to existing methods.

Conclusions:

• tascCODA is a valuable tool for generative modeling of compositional data.
• Facilitates the analysis of microbial and cell population compositional changes.
• Enhances understanding of complex biological systems through improved statistical analysis.