|
Title
|
|
|
|
Machine learning for decision support in adaptive immunology
| |
|
Author
|
|
|
|
| |
|
Abstract
|
|
|
|
| Our body has developed several defense mechanisms to remain in a healthy state. Collectively, these mechanisms are referred to as our immune system. This system can be largely decomposed into two major components: the innate immune system and the adaptive immune system. The importance of understanding the immune system cannot be overstated with regards to human health as it is implicated in infectious diseases, in cancer and in autoimmune disease and understanding it is instrumental to many of our solutions to these health threats. However, due to the complexity of this system, it is far from trivial to understand its intricacies and to generate new insights. During the past decades, computer algorithms have been developed to allow computers to derive patterns and learn from data without any human assistance. Specifically, the field of machine learning aims to develop algorithms that can learn and generalize from data without human intervention. These algorithms are able to discern complex patterns from data at a level that is impossible for humans. In this thesis, we aimed to apply several of such machine learning methods to tackle immunological or immunology-related questions. We were able to demonstrate the possibility of predicting which T cell receptors are capable of binding an epitope within the context of two HIV-derived, HLA-B*08 restricted epitopes. In addition, we show that it is possible to predict the cytomegalovirus serostatus based on the presence of specific T cell receptor sequences present on CD4+ memory T cells. Also for the cytomegalovirus virus, we found several risk factors for the reactivation of the virus in kidney transplant patients and created a classification model to predict this reactivation. Finally, we were able to relate early changes in gene expression to long term vaccination outcomes for both a hepatitis B vaccine and a measles-mumps-rubella booster vaccine. Overall, we show that machine learning methods are able to generate new immunological insights and that it is possible to create sensible, high-performing models that can support both investigators and clinicians working in immunology-related fields. |
| |
|
Language
|
|
|
|
English
| |
|
Publication
|
|
|
|
Antwerpen
:
University of Antwerp, Faculty of Science, Department of Mathematics and Computer Science
,
2019
| |
|
Volume/pages
|
|
|
|
187 p.
| |
|
Note
|
|
|
|
:
Laukens, Kris [Supervisor]
:
Suls, Arvid [Supervisor]
| |
|
Full text (open access)
|
|
|
|
| |
|