COZINE

COZINE infers microbial networks from compositional, zero-inflated high-throughput microbiome abundance data to estimate sparse conditional dependencies among taxa.

Key Features:

• Multivariate Hurdle Model: Employs a multivariate Hurdle model to jointly model binary presence/absence indicators and continuous abundance values for taxa.
• Sparse Network Inference: Estimates a sparse set of conditional dependencies to emphasize significant associations within microbial communities.
• Compositional Data Handling: Accounts for the compositional nature of relative abundance data constrained by fixed sample sums.
• Zero-Inflation Management: Models zero inflation arising from taxa absence across subjects by incorporating binary and continuous components.
• Computational Scalability: Maintains computational efficiency suitable for large-scale datasets from high-throughput sequencing studies.

Scientific Applications:

• Ecological and medical microbiome research: Infers community interaction structures and association networks in ecological and clinical microbiome datasets.
• Oral microbiome in leukemic patients: Applied to analyze oral microbiome networks in a cohort of leukemic patients to uncover associations relevant to disease progression or treatment outcomes.

Methodology:

Uses a multivariate Hurdle model to infer conditional dependencies by jointly modeling presence/absence indicators and continuous abundances, and evaluates performance via simulations of diverse microbial relationships with comparisons to existing approaches.