PhD defence by Birkir Reynisson

On 20 May 2021, Birkir Reynisson will defend his PhD thesis: Integration of multi-allele MHC II eluted ligands for robust prediction of CD4 epitopes

Time: 09:00
Place: Zoom, register for the defence here: https://dtudk.zoom.us/meeting/register/u5cod-qoqjksHNNmFXz1ofgFtbeIHXyoXeUr

Principal supervisor: Professor Morten Nielsen
Co-supervisor: Associate Professor Paolo Marcatili

Examiners:
Associate Professor Henrik Nielsen, DTU Health Tech
Professor Anthony Wayne Purcell, Monash University
Director in Genomic Immuno-Oncology Jens Vindahl Kringelum, Evaxion Biotech

Chairperson at defence:

Associate Professor Gisle Alberg Vestergaard

Abstract:

T helper cells play the role of an orchestrator in the immune system, controlling the activation and development of cells that target pathogens and cancer precisely and can destroy them efficiently. T helper cells' activation depends on recognizing foreign peptides bound to Major Histocompatibility Complex class II (MHC II) on the surface of cells presenting their environment and internal state. Therefore, learning the rules that govern peptide MHC II presentation is a critical step in our understanding of immune activation. Prediction models, guided by experimental measurements of peptide binding preferences, provide valuable insights into these processes and serve as tools for research and clinical applications.

In this thesis, we present work on training artificial neural networks to predict peptides that activate T helper cells, epitopes, by learning the binding preferences of MHC II from mixed data types. We introduce an algorithm, NNAlign_MA, that solves the problem of learning signals from complex peptide mixtures. This vastly expanded the available data to learn from, which lead us to develop the next version of a broadly applicable MHC II peptide binding predictor, NetMHCIIpan-4.0. This framework proved flexible, as with a similar approach applied to MHC II peptide data from cattle, we developed the first broadly applicable MHC II predictor for cattle: NetBoLAIIpan-1.0. Finally, critically comparing NetMHCIIpan-4.0 prediction with experiment, we find overlap but also disagreement. We analyze these regions of disagreement with the hypothesis that they may hold insight into the process of antigen presentation.

Learning what governs immune recognition would have a vast clinical impact on a myriad of diseases, including infection, auto-immunity, and cancer. MHC II predictors also have applications in protein drug development, reducing their immunogenicity and prolonging their usability.
 

Time

Thu 20 May 21
9:00 - 12:00

Organizer

DTU Sundhedsteknologi

Where

Zoom