Leveraging Interpretable Machine Learning for Multiomic and Clinical Data Integration in Biomarker Discovery for Precision Medicine - Dissertation

Aug 4, 2026

Upload by Quinn

Precision medicine enables tailored treatments based on individual molecular, environmental, and lifestyle factors, improving outcomes and reducing adverse effects. Persistent barriers include underused data modalities, data sparsity, limited interpretability, and the need for advanced computational tools to handle high-dimensional datasets. This thesis develops methodological advances spanning preclinical to clinical research. It enhances biomarker discovery via ancestry-informed stratification in high-throughput cancer cell drug screens, delivers an end-to-end interpretable framework integrating multiomic data for observational metabolic studies, and uses pretrained large language models to build interpretable prognostic models from longitudinal EHRs for type 2 diabetes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages171

Reading Time7 h 11 min

AccessFree to download

Abstract

Acknowledgements

Leveraging Interpretable Machine Learning for Multiomic and Clinical Data Integration in Biomarker Discovery for Precision Medicine Dissertation der Fakult¨at f¨ur Biologie Ludwig-Maximilians-Universit¨at M¨unchen Phong Ba Hung Nguyen Munich, 2025 Leveraging Interpretable Machine Learning for Multiomic and Clinical Data Integration in Biomarker Discovery for Precision Medicine Dissertation der Fakult¨at f¨ur Biologie der Ludwig-Maximilians-Universit¨at M¨unchen Phong Ba Hung Nguyen M¨unchen, 2025 Diese Dissertation wurde angefertigt unter der Leitung von A. Prof. Michael Menden im Bereich von Fakult¨at f¨ur Biologie an der Ludwig-Maximilians-Universit¨at M¨unchen Erstgutachter/in: A. Prof. Michael Menden Zweitgutachter/in: Prof. Dr. Wolfgang Enard Tag der Abgabe: 11.09.2024 Tag der m¨undlichen Pr¨ufung: 12.03.2025 Erkl¨arung Ich versichere hiermit an Eides statt, dass meine Dissertation selbstst¨andig und ohne unerlaubte Hilfsmittel angefertigt worden ist. Die vorliegende Dissertation wurde weder ganz noch teilweise bei einer anderen Pr¨ufungskommission vorgelegt. Ich habe noch zu keine fr¨uheren Zeitpunkt versucht, eine Dissertation einzureichen oder an einer Doktorpr¨ufung teilzunehmen. M¨unchen, den June 18, 2025 Phong Nguyen Abstract Precision medicine represents a pivotal advancement in healthcare, offering the potential to tailor treatments based on individual molecular, environmental, and lifestyle factors, thus improving patient outcomes and reducing adverse effects. However, significant challenges remain, including the underutilization of different modalities of data, data sparsity, issues with interpretability, and the need for sophisticated computational methods to analyze high-dimensional datasets. To address these challenges, this thesis introduces multiple advancements in methodology for various aspects of precision medicine, from preclinical to clinical studies. For preclinical studies such as drug high throughput screens of cancer cell lines, the stratification with inferred ancestry information enhanced biomarker discovery, demonstrating improved identification of drug response biomarkers. In addition, for observational clinical studies, integration of multiomic data within a biologically interpretable framework, providing an end-to-end comprehensive and transparent machine learning approach to biomarker discovery for complex metabolic diseases. Furthermore, for more complex data such as clinical longitudinal electronic health records, the utilization of pretrained large language models to develop an interpretable prognostic model for type 2 diabetes offered valuable insights into disease progression and patient management. Together, these proposed methods highlight the transformative potential of integrating advanced machine learning techniques and diverse data types to advance biomarker discovery in multiple aspects of precision medicine. Insights into disease mechanisms and actionable biomarkers discovered from these studies serve as valuable resources to help translate both biomedical research and healthcare practice and eventually benefit patients. Acknowledgements I would like to express my deep gratitude to several people who have helped me through this long journey. First and foremost, I would like to thank my supervisor, A/Prof. Dr. Michael Menden, who has given me the opportunity to do a PhD, guided me through the earliest time of my research journey and supported me throughout my development. In addition, I am grateful for my mentors and collaborators, Prof. Dr. Wolfgang Enard, Prof. Dr. Christian Herder, [Prof. Dr. med. Reinhard Holl](Prof. Dr. med. Reinhard Holl), Prof. Dr. Annette Peters, Dr. Sebastian Vosberg, Dr. Francesco Paolo Casale, for their invaluable feedback and support to make the projects happen. Furthermore, my PhD journey would not be fulfilled without the presence and assistance from the fellow researchers in my lab, includi

FAQs

What key challenge motivates this thesis in precision medicine?

The work targets gaps in precision medicine caused by underutilization of data modalities, data sparsity, interpretability issues, and the need for computational methods suited to high-dimensional datasets.

How does the thesis improve biomarker discovery in preclinical drug response studies?

It stratifies cancer cell lines using inferred ancestry information, which improves the identification of drug response biomarkers in high-throughput drug screens.

What approach is used for observational clinical studies involving multiomic data?

It integrates multiomic data within a biologically interpretable, end-to-end machine learning framework to support transparent biomarker discovery for complex metabolic diseases.

How are longitudinal electronic health records used in the study of type 2 diabetes?

The thesis employs pretrained large language models to develop an interpretable prognostic model, providing insights into disease progression and patient management.

Aug 4, 2026

Upload by Quinn