270978 Development of Computational Methods to Support De Novo Antibody Design

Monday, October 29, 2012: 8:48 AM
Fayette (Westin )
Robert J. Pantazes, Deparment of Chemical Engineering, The Pennsylvania State University, University Park, PA and Costas D. Maranas, Department of Chemical Engineering, The Pennsylvania State University, University Park, PA

The de novo design of a novel protein with a particular function remains a formidable challenge with only isolated and hard-to-repeat successes to date.  Due to their many structurally conserved features, antibodies are a family of proteins more amenable to predictable rational design. Previously, we have developed the Optimal Complementarity Determining Regions (OptCDR) method to design antigen binding sites targeted against any specified epitope.  OptCDR functions by selecting optimal combinations of backbone structures that are most likely to have favorable interactions with an antigen and then filling in their amino acids during a computational affinity maturation step.

Here we will present how we are moving beyond designing just the antigen recognition sites to the design of fully-human antibody variable domains. In this effort, antibody sequences, genes, structures, and standards are imported from the IMGT database [http://www.imgt.org]. The immune system naturally generates billions of antibodies from only hundreds of genes using VDJ recombination, where random variable (V), diversity (D), and joining (J) genes, each of which correspond to a specific region of antibody variable domain structure, are combined to create an antibody.  Using a clustering procedure we mapped the observed structural diversity onto less than 1,000 structures that span three regions, analogous to the V, D, and J genes.  The recombination of this collection of structures spans a potential diversity of over 6*10^9 unique human antibodies. We find that using this structural description we can predict the experimentally resolved structures with an average rmsd of 1.28 Å. Subsequently, we analyzed (i) the broadly neutralizing anti-influenza antibody CH65 and (ii) anti-HIV antibody 4E10 to provide starting points for the design of affinity-matured antibody libraries and examine the frequency and distribution of mutations during the affinity maturation process. Ultimately, the goal is to use the developed database within a computational method for the de novo design fully human antibody variable domains.

Extended Abstract: File Not Uploaded