Development of computational methodologies for antibody design

<p>Antibodies are proteins of the adapative immune system. Structural diversity in an antibody's two variable domains, V<sub>H</sub> and V<sub>L</sub>, allow it to bind almost any molecule with high affinity and specificity. This thesis focuses on characterising su...

Full description

Bibliographic Details
Main Author: Leem, J
Other Authors: Deane, C
Format: Thesis
Published: 2016
Description
Summary:<p>Antibodies are proteins of the adapative immune system. Structural diversity in an antibody's two variable domains, V<sub>H</sub> and V<sub>L</sub>, allow it to bind almost any molecule with high affinity and specificity. This thesis focuses on characterising such variations to develop computational tools for antibody design, with a particular interest toward engineering antibodies as therapeutics. </p> <p>First, we describe a method to predict the binding affinities of antibody–antigen interactions. Using the contacts at the antibody–antigen interface, we show promising results, but the performance is too poor for extensive design applications. Since several factors can influence antibody binding, we investigate V<sub>H</sub>–V<sub>L</sub> pairing, one of the largest sources of antibody structural variation. Based on our data, we describe a structure–based mechanism to describe V<sub>H</sub>–V<sub>L</sub> pairing. In particular, the high conservation of contacts at the V<sub>H</sub>–V<sub>L</sub>interface in over 6000 antibody sequences provides support for random V<sub>H</sub>–V<sub>L</sub> pairing. </p> <p>Following this analysis, we introduce our antibody modelling pipeline, ABodyBuilder. We demonstrate that ABodyBuilder can rapidly build accurate models, and is useful for mapping the antibody structural landscape from sequence. Furthermore, ABodyBuilder calculates the model's expected accuracy in order to help the decision–making process for users during antibody design. To complement ABodyBuilder's current setup, an antibody–specific rotamer library and side chain prediction algorithm are described. Although the maximum achievable accuracy is near 100%, the actual accuracy is closer to 80%, suggesting that the algorithm needs further refinement before full integration into the ABodyBuilder pipeline. </p> <p>Finally, we discuss how the tools presented in the thesis can be improved, and applied to other problems in computational antibody design. We also present an overview on the potential avenues for expanding the work herein.</p>