@phdthesis{01K6AN7CENRMG1M3SAKBXZN4A4,
  abstract     = {{DNA, RNA and proteins form the holy molecular trinity of biology, creating life as we know it. A complex flow of information regulated at many different steps ultimately enables linear DNA to create complex proteins, each dedicated to a specific task. These proteins have therefore found themselves an important subject of research, as a better understanding of these proteins, both in health and in disease, can provide new insights as to how our body functions, why things sometimes go wrong and how modern medicine can prevent or combat this. Due to this potential of protein studies, the importance of proteomics has skyrocketed and has led to many subfields each studying a smaller part of the whole leading to specific treatments and insights. 
One such subfield is immunopeptidomics which has shown great potential in identifying key components, i.e. epitopes, presented on the cell surface bound to MHC proteins. These epitopes enable our immune system to scan cells for pathogens, infections and mutations. It is through the identification of these epitopes that we can identify novel antigens that can be used in defeating viruses through vaccination or help our body to defeat cancer through cancer vaccinations or T-cell receptor technologies. It is therefore essential that we accurately identify as many of these epitopes as possible. However, currently epitope discovery is still hampered by the lack of dedicated bioinformatics software tailored to these immunopeptides.
Therefore, I dedicated my PhD towards the development of new algorithms that can further push the identification of such epitopes. Firstly, by developing new MS²PIP peak intensity prediction models that more accurately predict how these immunopeptides fragment in the mass spectrometer. Secondly, I have integrated these peak intensity prediction models together with retention time prediction models into a package called MS²Rescore. The orthogonal information from these prediction models creates a unique feature set for PSM rescoring, allowing MS²Rescore to significantly improve the amount of identified immunopeptides in a single experiment. To make MS²Rescore widely available and easy to use, I substantially refactored MS²Rescore as a third step in my PhD. This new, more modular package, alongside a graphical user interface and Python API has ensured that MS²Rescore can be easily adapted with new predictors, integrated into new pipelines and used by the wider community. This also enabled me to create a new version of MS²Rescore, called TIMS²Rescore. Indeed, as my fourth objective I fully adapted MS²Rescore to DDA-PASEF data, made popular through the emergence of timsTOF instruments. Through the integration of newly created peak intensity prediction models, ion mobility prediction models and a streamlined data throughput, I showed that TIMS²Rescore not only boosts immunopeptide identification rates but also improves plasma proteomics and metaproteomics data analysis. Lastly, to move beyond mass spectrometry-based identification of potential epitopes, I set out to create a new epitope predictor, MHC-3PO. By combining cutting-edge deep learning techniques such as large protein language models, graph neural networks and Siamese contrastive learning, I created an epitope predictor that is capable of generalising to unseen pMHC pairs and capable of predicting epitopes carrying PTMs, a field still largely understudied. Through MHC-3PO and its explainability I aim to further uncover this modified epitope universe going forward. 
In conclusion, this research has introduced innovative bioinformatics tools that push the boundaries of epitope discovery through mass spectrometry-based immunopeptidomics further enhancing our understanding of immune responses. These tools have shown great promise to advance immunotherapy applications, including cancer vaccines and T-cell receptor technologies.}},
  author       = {{Declercq, Arthur}},
  keywords     = {{Machine learning,bioinformatics,mass spectrometry,immunopeptidomics}},
  language     = {{eng}},
  pages        = {{I, 241}},
  publisher    = {{Ghent University. Faculty of Medicine and Health Sciences}},
  school       = {{Ghent University}},
  title        = {{Epitope Odyssey : data-driven tools to extend the known immunopeptide universe}},
  year         = {{2025}},
}