Back to search

University of Thessaly (UTH)

Algorithms for understanding the role of microbiome in human diseases

Abstract

dc:description

This dissertation contributes a complete computational system of microbial genomics and antimicrobial resistance (AMR) interpretation that brings together machine learning strategies, bioinformatics tool engineering, and clinical use to overcome significant challenges of human microbiome science, infectious disease diagnosis, and public health monitoring. Using new algorithm and analytical pipeline development, it identifies sound computational methods of uncovering microbial footprints of human health, defining AMR mechanisms of action, and predicting resistance phenotypes. The first component of this thesis employs machine learning methodology to identify microbial signatures within healthy human tissues and therefore challenged conventional notion about the sterility of human organs. Analysis of 13,871 RNA-seq samples from 28 tissues within the GTEx consortium, processed using AGAMEMNON for metatracnscriptomics quantification, revealed tissue-specific microbial signatures within eight tissues that are resilient to in silico contamination using gradient boosting models. In silico contamination analysis validated the robustness of the identified signatures against diverse possible sources of contamination. Independent validation using samples of living tissues confirmed that the heart, colon, and liver have unique low-biomass microbiomes with prominent signatures like Sporisorium graminicola within cardiac tissue, the gram-negative bacterium Bartonella machadoae within liver tissue, and Flavonifractor plautii within colon tissue. These results suggest non-random distributions of microbial colonization and challenge widely held assumptions about tissue sterility. The second component centers on training machine learning models designed to predict antimicrobial resistance phenotypes from genomic data of ESKAPEE pathogens. ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.), are known culprits of life-threatening hospital-acquired infections and are a global health concern because of their ability to evade widely used antibiotics. By using a carefully curated dataset of 18,916 genome assemblies and corresponding antibiograms of 40 antibiotics, the use of k-mer-based feature engineering for all known AMR genes, their upstream promoters, and rRNA sequences allowed the Random Forest and XGBoost classifiers to attain recall and F1-scores of higher than 90% on the vast majority of antibiotics. These models performed higher performance than rule-based tools, which predict the phenotype by only identifying AMR genes or point mutations. Further, SHAP analysis made interpretable observations on the genomic features underlying resistance calls. Clinical validation by metagenomic sequencing of 36 positive for bacteria, blood culture samples demonstrated rapid in silico antibiogram generation within 24-48 hours using the created models, substantially faster than the performance of conventional culture-based approaches. The third component of this thesis introduces AmrProfiler, a comprehensive web tool developed to identify antimicrobial resistance (AMR) genes, point mutations and ribosomal RNA gene variations within nearly 18,000 species of bacteria. Through data aggregation and complementation of information from authenticated databases (CARD, ResFinder, Reference Gene Catalog), AmrProfiler produces a holistic web tool for studying antimicrobial resistance, by identifying AMR genes, mutations in core genes that are associated with resistance and the first systematic investigation of rRNA gene mutations, enabling complete identification of resistance determinants with user-specifiable thresholds. It is made up of three expert modules—the acquired AMR genes, core gene mutations and rRNA gene analysis—which yield unprecedented depth within the delineation of AMR. Testing against a series of ESKAPEE pathogens assemblies verified its unparalleled effectiveness when compared with available tools (such as Resfinder, Card and AmrFinderPlus) correctly identifying all previously known resistance markers and detecting other mechanisms not recognized by tools currently available. Lastly, its intuitive easy-to-use interface make available for microbiologists without a deep knowledge of bioinformatics, an in depth AMR analysis.Collectively, this dissertation contributes to the science of computational microbiology by revealing new insights of human tissue microbiomes, providing valid tools of AMR characterization, and demonstrating the clinical significance of genomics-based resistance machine learning prediction. The integrated computational workflows herein, from metatranscriptomics interpretation and curation of databases to construction of machine learning models and clinical validation—offer reproducible platforms of microbiome research, AMR surveillance, and precision diagnosis. This thesis brings essential computational infrastructure toward understanding microbial communities and data-driven therapeutic decision-making.

Degree

thesis:*
Grantor dc:publisher
University of Thessaly (UTH)
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Skoulakis, Anargyros
  • Σκουλάκης, Ανάργυρος

Subjects

dc:subject × 22

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/60136

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Skoulakis, Anargyros; Σκουλάκης, Ανάργυρος. Algorithms for understanding the role of microbiome in human diseases. University of Thessaly (UTH), 2025. http://hdl.handle.net/10442/hedi/60136