Generative modelling and transformer-based sequence learning for microbiome analysis

Simple item page
dc.check.date2029-12-31
dc.check.infoControlled Access
dc.contributor.advisorClaesson, Marcus
dc.contributor.advisorTemko, Andriy
dc.contributor.authorButler, James C.en
dc.date.accessioned2026-10-01T07:25:15Z
dc.date.available2026-10-01T07:25:15Z
dc.date.issued2026-03-31
dc.date.submitted2026-03-31
dc.descriptionControlled Access
dc.description.abstractTwo decades of research have established the significant role of the microbiome in human physiology and its association with a wide range of diseases. However, characterizing the relationship between the microbiome and host phenotype remains challenging due to the difficulty of obtaining high-quality samples and the complex statistical properties of microbiome data, including high dimensionality, sparsity, and compositional structure. As sequencing technologies generate increasingly large and high-resolution datasets, there is a growing need for analytical frameworks capable of identifying disease-relevant signals while preserving interpretability. This thesis investigates recent advances in generative modelling and transformer-based architectures for microbiome analysis, with a focus on improving disease classification and assessing their potential to inform our understanding of the human microbiome. The first chapter provides an overview of generative artificial intelligence and transformer-based methods in the context of microbiome analysis. Chapter 2 introduces MicroDOSE, a hybrid generative framework that combines adversarial and diffusion-based modelling to synthesize biologically realistic microbiome samples, and evaluates the extent to which synthetic data can augment limited training cohorts while preserving key ecological characteristics. Chapter 3 presents LaMe Diffusion, which applies diffusion modelling in latent spaces derived from deep autoencoder representations of microbiome data and shows that classification performance is highly sensitive to representation quality, while positioning diffusion as a strategy for improving the stability of learned latent structure. Chapter 4 develops Meta-STOIC, a transformer-based architecture for disease classification directly from metagenomic sequencing reads, and evaluates the feasibility of predicting disease from sequence-level representations without reliance on predefined taxonomic or functional features. The thesis concludes by outlining future directions for building unified models that can capture the complexity and context-dependent behavior of microbial ecosystems, and help us better understand microbiome–host interactions.
dc.description.statusNot peer revieweden
dc.description.versionAccepted Versionen
dc.format.mimetypeapplication/pdfen
dc.identifier.citationButler, J. C. C. 2026. Generative modelling and transformer-based sequence learning for microbiome analysis. PhD Thesis, University College Cork.
dc.identifier.endpage193
dc.identifier.urihttps://hdl.handle.net/10468/19387
dc.language.isoen
dc.publisherUniversity College Corken
dc.rights© 2026, James Butler.
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/
dc.subjectGenerative artificial intelligence
dc.subjectMicrobiome
dc.subjectDiffusion
dc.subjectGenerative adversarial networks
dc.subjectTransformers
dc.subjectLarge language models
dc.titleGenerative modelling and transformer-based sequence learning for microbiome analysis
dc.typeDoctoral thesisen
dc.type.qualificationlevelDoctoralen
dc.type.qualificationnamePhD - Doctor of Philosophyen
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
ButlerJC_PhD2026.pdf
Size:
45.55 MB
Format:
Adobe Portable Document Format
Description:
Full Text E-thesis
License bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
license.txt
Size:
5.2 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
Research Theses
APC Microbiome Ireland - Doctoral Theses
College of Medicine and Health - Doctoral Theses
College of Science, Engineering and Food Science - Doctoral Theses
Microbiology - Doctoral Theses