Next-generation antibody modeling
Weitzner, Brian D
MetadataShow full item record
Antibodies are important immunological molecules that can bind a diverse array of foreign molecules. The genetic mechanism that gives rise to antibodies and many antibody sequences is known, but only by studying three-dimensional structures of antibodies and antibody–antigen complexes can we reveal immunological mechanisms and provide a starting point for developing rationally designed antibodies. With the advent of high-throughput sequencing technologies, the gap between the number of sequences and structures is widening, demanding accurate antibody modeling methods. Our previously developed method, RosettaAntibody, served as a starting point for antibody structure prediction. In this dissertation, I detail my work assessing the predictive power of RosettaAntibody, and the development and testing of new methods to address its weaknesses. First, I describe an effort to assess the accuracy of RosettaAntibody on a set of unpublished crystal structures. This challenge enabled us to combine manual and automated methods for selecting models and compare RosettaAntibody to other antibody modeling methods. The most challenging aspect of structure prediction in this assessment proved to be modeling the third complementarity determining region loop on the heavy chain (CDR H3). Next I detail my work in studying CDR H3 loops to uncover why a vast majority of them contain a kink at the loop's C-terminus. Part of this work involved searching the Protein Data Bank (PDB) for structures with a similar geometry of the amino acid residues at the base of the loop, leading to a set of CDR H3-like loops from non-antibody proteins. With a clearer understanding of CDR H3 loop structures and the most detailed description of the kink to date, I developed a new loop modeling routine that utilizes this information to restrict the geometry of the loop to be kinked, resulting in an improvement in the weakest aspect of antibody structure prediction. In summary, the structure prediction methods I have developed and structural analyses I have performed provide a means to begin to address the widening sequence–structure gap. Additionally, these methods can be used to perform structural analysis in the development of rationally designed antibodies.