Controlled Access. Restriction lift date: 2027-06-30
Computational approaches to gut microbiome analysis: colonization resistance, disease-driven network reorganisation, and hierarchical taxonomic classification
Loading...
Files
Date
2025-12-31
Authors
Kaka Bra, Kardokh
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
The gut microbiome represents a critical determinant of human health, yet fundamental challenges persist in accurately characterising microbial communities, understanding their protective functions against pathogens, and elucidating how functional relationships between microbes and metabolites are disrupted in disease. This thesis addresses these interconnected challenges through the development and application of computational approaches spanning host-pathogen-microbiome dynamics, systems-level analysis of microbiome-metabolome networks, and taxonomic classification methodology.
The first contribution is a computational analysis of colonisation resistance against Listeria monocytogenes in defined and complex microbiota mouse models. Comparison of the 12-member Oligo-MM12 consortium with specific pathogen-free (SPF) communities reveals that microbiota complexity shapes infection dynamics in unexpected ways. A striking temporal reversal was observed: Oligo-MM12 mice showed significantly lower Listeria burden at day 1 but significantly higher burden by day 3 compared to SPF mice, suggesting that functional redundancy conferred by community diversity is essential for sustained rather than immediate protection. The identification of Akkermansia, Blautia, and Prevotellamassilia as infection-responsive genera provides targets for mechanistic investigation. Parallel analysis of in vivo and ex vivo fermentation systems revealed that host factors drive most observed taxonomic perturbations during infection, while microbiome-intrinsic mechanisms independently contribute to pathogen control.
The second contribution is a network-based framework for investigating microbiome-metabolome disruption across six human disease cohorts encompassing colorectal cancer, gastric cancer, inflammatory bowel disease, irritable bowel syndrome, and end-stage renal disease. Disease-stratified bipartite network construction reveals that disease involves not merely compositional changes but fundamental network reorganisation. Healthy networks harbour substantially more conserved keystone species than disease networks, with a 50.9% reduction in cross-study conserved keystones during disease. Approximately ten bacterial species—predominantly obligate anaerobes from Lachnospiraceae, Oscillospiraceae, and Ruminococcaceae families—showed consistent depletion across three or more independent disease cohorts, representing universal signatures of gut ecosystem disruption. However, network topology changes were disease-specific, with some conditions showing reduced connectivity and others showing increased density, indicating that different pathophysiological mechanisms drive microbiome-metabolome reorganisation in distinct disease contexts.
The third contribution is Taxonova, a hierarchical deep learning framework for species-level taxonomic classification from amplicon sequences. By processing taxonomic levels sequentially from domain to species with multi-head attention mechanisms, and employing a hierarchical focal loss function to address class imbalance in reference databases, Taxonova achieves F-measures of 0.86–0.99 across amplicon lengths ranging from 250 bp V4 regions to 4,900 bp full-length ribosomal RNA operons. The framework demonstrates 2–10-fold reduction in abundance estimation errors compared to existing classifiers and up to 4.2-fold higher concordance with paired shotgun metagenomic data in both high-complexity gut and low-complexity vaginal microbiomes. These results demonstrate that the resolution gap between amplicon and shotgun sequencing is partially algorithmic, with advanced classification methods extracting substantially more information from cost-effective amplicon approaches.
Synthesising across these investigations, several integrative themes emerge: species-level resolution is essential for identifying functionally important taxa; obligate anaerobic bacteria appear consistently as health-promoting taxa vulnerable to disruption across diverse pathological contexts; and complexity and functional redundancy confer ecosystem resilience. The principal contributions of this thesis are methodological frameworks for hierarchical taxonomic classification and disease-stratified network analysis, biological insights into systems-level organisation of the gut ecosystem in health and disease, and translational targets including keystone species representing candidates for further mechanistic investigation and, subject to species-level validation in human-relevant models, potential targets for microbiome-based therapeutic development and network-based biomarkers for diagnostic development.
Description
Controlled Access
Keywords
Microbiome , Colonisation resistance , Listeria , Machine learning , Multi-omics
Citation
Kaka Bra, K. 2025. Computational approaches to gut microbiome analysis: colonization resistance, disease-driven network reorganisation, and hierarchical taxonomic classification. PhD Thesis, University College Cork.