Identification of keystone-species critical for the systems diversity, stability or human health

Microorganisms have inhabited planet Earth for billions of years, and have evolved to colonize virtually every possible environment, including the human body. Hundreds of different bacterial species reside in the human gastrointestinal tract, and these are part of the human gut microbiota. A healthy microbiota can be beneficial for the human host as it performs many relevant functions, including the metabolism of dietary components and the training of the immune system. In healthy individuals, the microbiota exhibits a large diversity and its composition is rather stable. Abnormal shifts in the microbiota composition have been linked with the onset and progression of various diseases. Studying these shifts can result in the discovery of diagnostic biomarkers and lead to possible disease treatments.

In this thesis, we explore the within-species diversity of bacteria in the human gut and the relation between the gut microbiome and health. Bioinformatics methods are applied on high-throughput DNA sequencing data to investigate (i) the associations between species and subspecies to the health status of their host, and (ii) the influence of subspecies on other species. The techniques we apply include abundance estimation of individual species, known as species profiling, and the discovery of genetic differences in bacterial genomes at the level of individual nucleotides, a process known as single-nucleotide variant calling.

In Manuscript 1, we address the co-existence among microorganisms in the human gut. We find that many subspecies influence the abundance of other species and unveil that those interactions may be explained by the genomic differences among the subspecies. This study provides relevant insights into the evolution of bacterial species and reveals that different subspecies can play distinct roles in the modulation of the microbial composition. This underlines, once more, that the human gut microbiome is a highly complex ecosystem.

Non-alcoholic fatty liver disease (NAFLD) is a chronic disorder of the liver with a reported global prevalence of about 25 %, yet the awareness of NAFLD is low. While the exact mechanisms of the disease are yet to be determined, it has been suggested that imbalances of the gut microbiota and its functions are potentially involved in the disease development. In Manuscript 2 we employ machine learning (ML) to predict if individuals NAFLD-free at the time of sampling will have developed NAFLD after 4-years. We demonstrate that our prognostic ML model allows the identification of potential biomarkers for NAFLD and
generalizes well on external cohorts.

Finally, we present a novel method in Manuscript 3 that exploits the associations found between clades of subspecies and human health to engineer features suitable for the training of ML models. This is based on the concept that genetic variation within bacterial species may result in alterations in pathogenicity. We evaluate the approach in a dataset for inflammatory bowel disease (IBD) subjects and controls and show that the integration of such associations as engineered features provides useful insights about the relations between specific clades and the disease, and that these can be relevant in ML disease predictors.