Abstract
The intestinal microbiome is the collective set of bacteria and genes that inhabits the human gastrointestinal tract. The advent of high-throughput sequencing technologies in the last decade has allowed to correlate dynamics of this community with the health and disease, with implications for the development of infectious, inflammatory and neurobehavioral conditions. While one of the ultimate goals of microbiome sciences is to predict host phenotype or clinical outcome from microbiome colonization data, the high dimensionality of microbiome datasets, the non-linearity of the biological processes and the not-normality of sequencing data prevents the use of traditional regression based techniques to do this. In this dissertation we use and develop computational infrastructure that leverages random forest regression and classification to predict host phenotype from microbiome data from both experiments with mice and human clinical data. In the first study of this work, we applied our algorithms on how microbiome and clinical covariates predict Alzheimer’s dementia and other forms of dementia in nursing home elders. Our analysis not only was able to accurately classify these elders with respect to dementia status from both clinical parameters and abundance of microbial taxa and metabolites, but also allowed us to identify a host immune pathway underlying the relationship between the gut and the brain. In a second study, we applied our machine learning algorithms to determine to what extent microbiome dynamics post anti-Mycobacterium Tuberculosis treatment affect system host gene expression reprogramming. We show that even though anti-TB drugs caused profound microbiome changes even after two weeks of treatment, changes in peripheral gene expression could be most exclusively associated to the anti-tubercular activity of one of the drugs. Finally, in the third study of this work, we developed and applied our machine learning methods to assess the impact of specific host immune cells, peripheral regulatory T-cells on early microbiome development in mice. By mining both DNA sequencing and metabolomics data we showed that peripheral regulatory T-cells deficiency led to heightened type 2 immune responses triggered by microbial exposure, which disrupted the niche of border-dwelling bacteria early during colonization and that impaired peripheral regulatory T-cells generation led to pervasive changes in metabolite profiles.