Statistical Methods for Modeling Complex Dependency Structures in Zero-inflated Metagenomic Sequencing Data