sparseDOSSA

sparseDOSSA models and simulates microbial community profiles to generate realistic, parameterized synthetic microbiome datasets that preserve sparsity, zero-inflation, compositionality, sequencing depth, and known covariance structures for benchmarking microbiome analysis methods.

Key Features:

• Zero-inflated log-normal abundance model: Models microbial feature abundances using a zero-inflated log-normal distribution to capture sparsity and zero-inflation.
• Absolute cell count and sequencing depth components: Augments abundance modeling with components that account for absolute cell counts and sequencing depth.
• Interaction modeling: Incorporates microbe–microbe and microbe–environment interactions into simulated profiles.
• Compositionality and non-independence handling: Represents compositional constraints and non-independence among taxa in the simulated data.
• Parameterized simulations from empirical structures: Derives simulation parameters from empirical microbial population structures to produce realistic datasets.
• Known covariance structures: Generates synthetic datasets with fully known covariance structures between taxa and between taxa and phenotypes.
• Spiking of true associations: Enables spiking in true positive associations for controlled evaluation of methods.
• Unified benchmarking framework: Provides a unified system for evaluating and comparing microbiome analysis tools using realistic simulations.

Scientific Applications:

• Benchmarking microbiome methods: Evaluates and compares statistical and bioinformatics methods for microbiome analysis using realistic simulated data.
• Modeling human-associated microbial communities: Simulates human-associated microbiome profiles that reflect empirical population structures.
• Generating controlled ecological patterns: Produces synthetic populations with controlled ecological patterns for hypothesis testing.
• Power and association testing: Assesses power and false discovery through spiking true positive associations into simulations.
• Reproducing experimental workflows: Recreates full experimental workflows such as mouse microbiome feeding studies for validation of analysis pipelines.
• Validating statistical methods: Provides ground-truth datasets to validate statistical inference under sparsity, compositionality, and non-independence.

Methodology:

Models use a zero-inflated log-normal distribution augmented with components for absolute cell counts and sequencing depth, and incorporate microbe–microbe and microbe–environment interaction terms to produce parameterized simulations and synthetic datasets with known covariance structures, with options to spike in true associations derived from empirical microbial population structures.