Advanced Machine Learning Techniques for Biomarker Discovery and Disease Diagnosis Using Metabolomic Data

Omics technologies, and metabolomics in particular, are reshaping the biomedical landscape by providing comprehensive molecular snapshots of physiological and pathological states. These high-throughput approaches enable the identification of disease-related biomarkers and support the development of personalized interventions. Yet, the complexity, heterogeneity, and high dimensionality of omics data pose major analytical challenges, requiring advanced computational tools to fully unlock their potential. The overarching aim of this thesis is to explore how artificial intelligence (AI) and machine learning (ML) can be applied to metabolomics to transform complex datasets into clinically meaningful insights. Specifically, this work seeks to advance precision medicine by enabling the discovery of disease-specific biomarkers, improving diagnostic accuracy, and monitoring treatment responses through the development and application of computational methods. The objectives of the thesis are threefold: (i) to develop ML models capable of differentiating between closely related diseases, exemplified by osteoarthritis and rheumatoid arthritis, using synovial fluid metabolomics; (ii) to explore genotype-associated metabolic alterations linked to schizophrenia risk through integrative metabolomics analysis; and (iii) to conduct a large-scale meta-analysis, within the Horizon Europe Better4U project, to identify metabolic biomarkers linked to obesity and functional food interventions across diverse cohorts. To achieve these aims, robust ML pipelines were designed that encompass preprocessing, dimensionality reduction, feature selection, predictive modeling, and explainable AI approaches, ensuring both interpretability and reproducibility. Collectively, the studies in this thesis illustrate how AI- and ML-driven metabolomics can address critical bottlenecks in biomedical data analysis and contribute to the long-term goal of more personalized, accurate, and data-driven healthcare.

Grantor dc:publisher

Università degli Studi di Cagliari

Year dc:date

2026

Author dc:creator

• KOPEĆ, KAROLINA KRYSTYNA

Contributors dc:contributor

• FANOS, VASSILIOS
• DESSI', ANGELICA

• Settore MEDS-20/A - Pediatria generale e specialistica

Statement dc:rights

• info:eu-repo/semantics/embargoedAccess

Language dc:language

eng

Handle dc:identifier

https://hdl.handle.net/11584/475166

OAI identifier oai:identifier

oai:iris.unica.it:11584/475166