Computing and Information Systems - Theses
Permanent URI for this collection
Search Results
1 - 2 of 2
-
ItemMachine Learning Models for Vaccine Development and Immunotherapy( 2023-11)Therapeutic antibodies offer exceptional specificity and affinity to detect and eliminate antigens, making them valuable as therapeutics and in diagnostics. The antigen recognition and neutralisation is based on the efficient binding to epitopes, antigen regions recognised by antibodies that elicit an immune response. The identification and mapping of epitopes, however, are yet dependent on resource-intensive experimental techniques that do not scale adequately given the vast search space and diversity of antigens. Epitope identification and prioritisation is a cornerstone of immunotherapies, antibody design, and vaccine development. Consistent progress of computational approaches has been observed to improve in silico epitope prediction at scale, specifically driven by machine learning algorithms in the past decade. Yet, low predictive power and skewed data sets towards specific pathogens can still be observed. This thesis focused on better exploring publicly available experimental antibody-antigen data, improving modelling and identification of distinguishing epitope features that de- rive meaningful biological insights. On this basis, I have curated high-quality data from multiple resources, resulting in large scale and non-redundant epitope data sets. Besides, I proposed novel featurisation techniques grounded on graph-based approaches to model and discriminate epitopes from the remainder antigen surface, that were demonstrated to differentiate both classes. In addition, I have leveraged machine learning algorithms and data analysis for better predictive and explainable models, which have been translated and made available as easy-to-use web servers with Application Programming Interfaces for programmatic access and integration into Bioinformatics pipelines. By exploring these advanced computational methods, this thesis significantly contributes to improving the prediction of B-cell epitopes, leading to a better understanding of antibody targets, which I believe will facilitate the ongoing development of therapeutics and diagnostics.
-
ItemMachine Learning Models for Vaccine Development and Immunotherapy( 2023-11)Therapeutic antibodies offer exceptional specificity and affinity to detect and eliminate antigens, making them valuable as therapeutics and in diagnostics. The antigen recognition and neutralisation is based on the efficient binding to epitopes, antigen regions recognised by antibodies that elicit an immune response. The identification and mapping of epitopes, however, are yet dependent on resource-intensive experimental techniques that do not scale adequately given the vast search space and diversity of antigens. Epitope identification and prioritisation is a cornerstone of immunotherapies, antibody design, and vaccine development. Consistent progress of computational approaches has been observed to improve in silico epitope prediction at scale, specifically driven by machine learning algorithms in the past decade. Yet, low predictive power and skewed data sets towards specific pathogens can still be observed. This thesis focused on better exploring publicly available experimental antibody-antigen data, improving modelling and identification of distinguishing epitope features that derive meaningful biological insights. On this basis, I have curated high-quality data from multiple resources, resulting in large scale and non-redundant epitope data sets. Besides, I proposed novel featurisation techniques grounded on graph-based approaches to model and discriminate epitopes from the remainder antigen surface, that were demonstrated to differentiate both classes. In addition, I have leveraged machine learning algorithms and data analysis for better predictive and explainable models, which have been translated and made available as easy-to-use web servers with Application Programming Interfaces for programmatic access and integration into Bioinformatics pipelines. By exploring these advanced computational methods, this thesis significantly contributes to improving the prediction of B-cell epitopes, leading to a better understanding of antibody targets, which I believe will facilitate the ongoing development of therapeutics and diagnostics.