Machine learning approach for profiling human microbiome.

摘要

Understanding and characterizing the roles and variation of the microbial organisms living within the human host is the main goal of the international human microbial research community. These human flora has shown the ability to harvest otherwise inaccessible nutrients, synthesize vitamins and speed up the gut epithelial cell renewal process. Successful characterization of the microbial communities will have enormous impact towards improved human nutrition, immunity, and medical treatments. The advent of highly parallel Next Generation DNA sequencing technologies (NGS) makes it possible to determine the genomic content of not only individual organisms, but entire pools of co-existing species. Sequencing of microbial organisms collectively, is referred to as metagenomic and the comprehensive characterization of microbes within the human host is referred to as the "microbiome". Using advanced computational approaches, information about the microbial functions and community characterizations can be elucidated from those complex data.;The research in this area is spearheaded by the Human Microbiome Project (HMP) initiative at the National Institute of Health. The main goal of HMP is to generate resources that facilitate the extensive characterization of the human microbiota and investigation of their functions in human body. By combining both the NGS technologies and traditional approaches for functional analysis, HMP will provide the fundamentals for basic and advanced studies of microbial communities associated with human. The analysis of the deluge of information from the HMP and other similar efforts has opened a new field of microbiome research that is in severe need of computational tools suitable for many areas of data analysis. The general objective of this dissertation is to develop advanced computational tools specialized in the human microbiome research to characterize the microbial community based on sequence data, visualize the community patterns and discover significant correlations between microbial taxa and clinical diseases. These tools are developed using machine learning approaches that are appropriate for analyzing complex microbiome-related datasets due to their ability to extract underlying information from the large and noisy data produced in this research field. In this dissertation, machine learning tools were specifically developed for microbiome analysis. The suite of tools includes classification and clustering approaches coupled with an easy-to-use web interface made publicly available. It is hoped that this work will provide a foundation for further analysis of the microbiome.