Sequence-based and structure-based methods for studying the adaptive immune response
| dc.contributor.advisor | Kavraki, Lydia E. | en_US |
| dc.creator | Fasoulis, Romanos | en_US |
| dc.date.accessioned | 2025-01-16T19:33:29Z | en_US |
| dc.date.created | 2024-12 | en_US |
| dc.date.issued | 2024-08-23 | en_US |
| dc.date.submitted | December 2024 | en_US |
| dc.date.updated | 2025-01-16T19:33:29Z | en_US |
| dc.description.abstract | The adaptive immune system comprises various biological mechanisms that, in unison, protect an organism against various threats, such as pathogens, viral infections, and tumor cells. One of such mechanisms involves the binding of intracellular protein fragments called peptides to class-I Major Histocompability Complexes (MHCs). The formed peptide-MHC (pMHC) complex is presented to the surface of the cell, where it interacts with the T-cell receptor, an interaction that can elicit an immune response. Knowing which peptides bind to MHCs, which peptides are presented to the surface of the cell, and which peptides elicit an immune response is crucial for successful clinical applications and therapies. Due to the advent of mass spectrometry resulting in high-throughput generation of amino acid sequence-based pMHC binding data, amino acid sequence-based Machine Learning (ML) approaches have dominated the field, showing immense potential. At the same time however, it is known that the pMHC interaction is characterized by a strong structural component that is shown to be extremely important in fully explaining pMHC binding and peptide immunogenicity. This thesis presents methodologies that attend to both the amino acid sequence component and the structural component of the pMHC interaction. Focusing on the sequence component first, we present TLStab and TLImm, two ML-based tools that predict peptide kinetic stability and peptide immunogenicity respectively. Developed through adopting transfer learning methodologies, TLStab and TLImm outperform state-of-the-art approaches in their respective tasks. Next, focusing on the structural component, we present APE-Gen2.0, a new pMHC structural modeling tool. APE-Gen2.0 outperforms other approaches in the literature in regard to modeling accuracy. It also expands the pMHC structural modeling repertoire to peptides exhibiting post-translational modifications, as well as peptides that assume non-canonical geometries in the MHC binding cleft. Finally, we present RankMHC, a novel, Learning to Rank-based pMHC binding mode identification tool. RankMHC outperforms both classical protein-ligand scoring functions and pMHC-specific scoring functions in predicting the most representative peptide conformation among an ensemble of conformations. Overall, acknowledging the potential of both pMHC sequence and pMHC structure information, our work expands on both areas, through novel and effective computational contributions. | en_US |
| dc.embargo.lift | 2025-12-01 | en_US |
| dc.embargo.terms | 2025-12-01 | en_US |
| dc.format.mimetype | application/pdf | en_US |
| dc.identifier.uri | https://hdl.handle.net/1911/118168 | en_US |
| dc.language.iso | en | en_US |
| dc.subject | peptide-MHC | en_US |
| dc.subject | Machine learning | en_US |
| dc.subject | peptide-MHC binding affinity | en_US |
| dc.subject | peptide-MHC structural modeling | en_US |
| dc.title | Sequence-based and structure-based methods for studying the adaptive immune response | en_US |
| dc.type | Thesis | en_US |
| dc.type.material | Text | en_US |
| thesis.degree.department | Computer Science | en_US |
| thesis.degree.discipline | Computer Science | en_US |
| thesis.degree.grantor | Rice University | en_US |
| thesis.degree.level | Doctoral | en_US |
| thesis.degree.name | Doctor of Philosophy | en_US |