Machine Learning in Identification of Disease-Associated Microbiota

Abstract

Metagenomic studies of the microbiome community have revealed associations of the microbiome community to host disease state. The detection of these associations can rely on statistical analyses identifying differentially abundant taxa between diseased and healthy populations. Accurate prediction of the host phenotype from a metagenomic sample and identification of the associated microbial markers are important in understanding potential host-microbiome interactions related to disease initiation and progression. However, associations of individual microbes to a particular disease have shown contradictory results in past studies, possibly due to dynamic and complex natures of different microbes. To handle the complex nature of the microbiome, machine learning methods have begun being employed. Machine learning algorithms are a set of methods in which a model learns intrinsic patterns in data and use them to predict labels of data. In this chapter, we introduce the commonly used machine learning methods in metagenomic studies. We show readers how to use the currently available tools found in Python libraries. Our purpose is to demonstrate the proper training and analysis of machine learning models for microbiome researchers, who may not have experience in machine learning or Python programming.