Machine Learning for Biomedical Applications: From Clinical Complications to PCR Bias
Open access
Author
Date
2024Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
In an era where machine learning~(ML) are reshaping the landscape of biomedical research, this thesis presents a series of integrated studies that demonstrate the profound impact of ML in tackling complex challenges within related applications. The research undertaken in this thesis spans from improving predictive analytics in medical informatics to advancing data interpretation methods in biotechnological research, thus setting a trajectory for future innovation in these fields.
The first study investigates the critical area of pediatric sepsis, a major contributor to global childhood mortality~\citep{tan2019global}. Using a comprehensive dataset from major pediatric hospitals in Switzerland, including data on patients with confirmed bacterial infections, this research leverages machine learning to predict the progression of multiple organ dysfunction syndrome in pediatric sepsis patients. The developed machine learning models exhibit high accuracy in predicting clinical outcomes and demonstrate transferability when applied to another international patient database. This highlights the potential of ML as a valuable tool in assisting clinicians to make timely, informed and life-saving decisions.
Delving into the intricacies of multisource datasets, the second study introduces Joint Multidimensional Scaling~(Joint MDS), a novel technique for unsupervised manifold alignment. By harmonizing datasets from different domains into an aligned low-dimensional feature space, Joint MDS empowers researchers to uncover latent correspondences without the need for explicit data instance matching. This method's adaptability is exemplified through its application to a range of biological datasets, from single-cell multi-omics data to protein-protein interaction networks. It illustrates the transformative power of ML in facilitating a deeper understanding of complex, high-dimensional biological data, thereby advancing the goal of precision medicine.
The third study in this thesis tackles a critical challenge in biotechnology: the inherent bias in the Polymerase Chain Reaction~(PCR), a cornerstone technique in genomic research. This study introduces a comprehensive framework, employing advanced machine learning techniques coupled with both theoretical and empirical methodologies, to identify and mitigate PCR bias in DNA sequencing experiments. This framework not only sheds light on the complex dynamics of PCR bias but also paves the way for improving the effectiveness of PCR-based genomic studies. The implications of this research are far-reaching, offering potential enhancements in various applications, including disease diagnostics and the development of therapeutic strategies.
Overall, in this thesis we aim to illustrate the remarkable potential of machine learning to not only improve predictive analytics in medical informatics but also to progress data interpretation techniques in biotechnological studies, by utilizing these studies to weave a narrative. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000674005Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichSubject
Machine Learning; Healthcare; Biomedical applicationsOrganisational unit
02060 - Dep. Biosysteme / Dep. of Biosystems Science and Eng.
Funding
813533 - Machine Learning Frontiers in Precision Medicine (EC)
More
Show all metadata
ETH Bibliography
yes
Altmetrics