B1914
Title: Model-based bicluster algorithm for microbiome data
Authors: Zhili Qiao - Iowa State University (United States)
Peng Liu - Iowa State University (United States) [presenting]
Abstract: With the advancement of next-sequencing technologies, huge amounts of microbiome data have become available. Bicluster analysis is a tool to quantitively explore the relationships between microbial samples and between features simultaneously, and aims to reveal the interactions between microbial sub-communities. Due to compositionality and sparsity, it is challenging to conduct bicluster analysis with microbiome data. We propose a Dirichlet-Multinomial (DM) model-based checkerboard biclustering method to cluster microbiome features and samples simultaneously. This method assumes a mixture of DM distributions across microbiome samples, and uses a combination of the Expectation-Maximization algorithm and coordinate descent algorithm to solve for parameter estimates and achieve biclustering results. Simulation studies under a variety of settings show that our proposed method outperforms alternative methods. Application to a real dataset demonstrates the effectiveness of our proposed method and provides interesting biological findings.
