A Statistical Framework for Data Purification with Application to Microbiome Data Analysis

Identification of disease-associated microbial species is of great biological and clinical interest. However, this investigation still remains challenging due to heterogeneity in microbial composition between individuals, data quality issues, and complex relationships among species. In this paper, we propose a novel data purification algorithm that allows the elimination of noise observations, which leads to increased statistical power to detect disease-associated microbial species. We illustrate the proposed algorithm using the metagenomic data generated from colorectal cancer patients.