TY - BOOK

T1 - Accelerating vaccine development througha deep probabilistic programming approachto protein structure prediction

AU - Thygesen, Christian Bahne

PY - 2023

Y1 - 2023

N2 - In this thesis, three manuscripts will be presented that focus on utilizing anovel approach to protein structure prediction, subsequently allowing theacceleration of vaccine development.The first manuscript presents a deep, probabilistic, and generative model oflocal protein structure. The proposed model represents a means of evaluatingthe possible conformations that small protein fragments adopt. The modelproduces fragment libraries at a quality on-par with state-of-the-art models,at a fraction of the run time, without the need for external information andthird-party tools to guide the library construction.In manuscript 2 I use this model of local protein structure to accelerate thevaccine design process. Vaccines typically induce an immune response throughthe combination of structural B cell epitopes and small linear T cell epitopes.In manuscript 2, I present an approach that uses the local protein structuremodel to modify the coronavirus spike protein through peptide grafting. Weshow that the model can adapt the spike protein of SARS-CoV-2, in a man-ner that preserves the important B-cell epitopes needed to induce an antibodyresponse, while enriching for T cell epitopes that can boost this response. Ishow that vaccine constructs designed using this model express at a higherlevel than those designed with a naive approach allowing only small modi-fications of the spike protein. The vaccine constructs are able to induce anantibody response against the wildtype in immunized mice, indicating properfolding of the modified protein construct.The model presented in manuscript 1 focuses only on inferring backbonedihedral angles. This focus on internal coordinates limit the ability to modelfragments that are larger than 9 amino acids. The third and final manuscriptpresents a means to alleviate this problem by introducing a novel multi-scaleapproach employing likelihoods over both internal coordinates as well as re-constructed 3D-coordinates. I show that this change improves the modelsperformance on short fragments while allowing modelling of longer proteinfragments as well.

AB - In this thesis, three manuscripts will be presented that focus on utilizing anovel approach to protein structure prediction, subsequently allowing theacceleration of vaccine development.The first manuscript presents a deep, probabilistic, and generative model oflocal protein structure. The proposed model represents a means of evaluatingthe possible conformations that small protein fragments adopt. The modelproduces fragment libraries at a quality on-par with state-of-the-art models,at a fraction of the run time, without the need for external information andthird-party tools to guide the library construction.In manuscript 2 I use this model of local protein structure to accelerate thevaccine design process. Vaccines typically induce an immune response throughthe combination of structural B cell epitopes and small linear T cell epitopes.In manuscript 2, I present an approach that uses the local protein structuremodel to modify the coronavirus spike protein through peptide grafting. Weshow that the model can adapt the spike protein of SARS-CoV-2, in a man-ner that preserves the important B-cell epitopes needed to induce an antibodyresponse, while enriching for T cell epitopes that can boost this response. Ishow that vaccine constructs designed using this model express at a higherlevel than those designed with a naive approach allowing only small modi-fications of the spike protein. The vaccine constructs are able to induce anantibody response against the wildtype in immunized mice, indicating properfolding of the modified protein construct.The model presented in manuscript 1 focuses only on inferring backbonedihedral angles. This focus on internal coordinates limit the ability to modelfragments that are larger than 9 amino acids. The third and final manuscriptpresents a means to alleviate this problem by introducing a novel multi-scaleapproach employing likelihoods over both internal coordinates as well as re-constructed 3D-coordinates. I show that this change improves the modelsperformance on short fragments while allowing modelling of longer proteinfragments as well.

M3 - Ph.D. thesis

BT - Accelerating vaccine development througha deep probabilistic programming approachto protein structure prediction

PB - Department of Computer Science, Faculty of Science, University of Copenhagen

ER -