Abstract:
Traditional trial-and-error based approaches to vaccine design have been remarkably successful. One of the major successes was the eradication of smallpox in the 1970s. However, there are still many diseases for which no viable vaccine could be found, HIV infection and cancer being among the most prominent examples. Here, new, rationally designed types of vaccines, as, e.g., epitope-based vaccines (EVs) are a promising alternative. Due to their manifold advantages and their applicability in personalized medicine EVs have recently been attracting significant interest. EVs make use of target-specific immunogenic peptides, i.e., epitopes, to trigger an immune response.
In this thesis we propose new approaches to the in silico design of EVs. Given a set of target antigens, the first step in EV design is the discovery of candidate epitopes. Computational approaches to epitope discovery comprise major histocompatibility complex (MHC) binding prediction and T-cell reactivity prediction. The key problem in MHC binding prediction is the lack of experimental data for the vast majority of known allelic MHC variants. We present two support vector machine (SVM)-based approaches to overcome this problem. The first approach improves the predictive power of SVMs for alleles with little experimental binding data. The second approach — for the first time — allows predictions for all known MHC variants by exploiting structural similarities between different MHC molecules. The key problem in T-cell reactivity prediction are the complex dependencies of T-cell reactivity on the host proteome. We present the first approach that takes these dependencies into account. Our method markedly outperforms previously proposed approaches, indicating the validity of our approach.
Due to regulatory, economic, and practical concerns only a small number of candidate epitopes can be included in the EV. Hence, it is crucial to identify the optimal set of peptides for a vaccine. We formulate the epitope selection problem within a mathematical framework based on integer linear programming. The resulting optimization problem can be solved efficiently and yields a provably optimal peptide combination. We can show that the method performs considerably better than existing solutions. Furthermore, the framework is very flexible and can easily handle additional criteria.
For EV delivery, the selected epitopes are commonly concatenated into a single polypeptide. Since an unfavorable epitope order can result in the degradation of the intended epitopes, optimal epitope assembly is critical for the success of the EV. We present a graph-theoretical formulation of this problem that allows the efficient determination of optimal epitope orders. Application of the presented EV design approaches to realistic vaccine design studies yields promising results.