(270c) De Novo Computational Design of Fully-Human Antibody Variable Domains

The de novo computational design of novel proteins remains a formidable, but essential, challenge with only rare successes to date. Antibodies are an excellent initial system to learn reliable de novo design principals, due to their many modular structural features and functions that are limited to binding and not catalysis. Previously we have developed and presented at AIChE the OptCDR method, for designing antibody complementarity determining regions, and a database of Modular Antibody Parts (MAPs) for reliably predicting antibody structures. We have built upon these successes to develop an Optimal Method of Antibody Variable region Engineering (OptMAVEn) to design fully human antibody variable domains from scratch to bind any specified antigen epitope.

The MAPs database, which can be used to predict antibody structures with an average all-atom RMSD of 1.900 angstroms, is an all-atom representation of portions of antibody structures. OptMAVEn begins by searching the MAPs database for the initial antibody that can best bind the specified antigen epitope. Once such an antibody is identified, a computational affinity maturation is carried out using our Iterative Protein Redesign & Optimization procedure, with restraints that ensure that all 9-mer sequences are fully human. If desired, a library of the most promising mutants can be identified once affinity maturation is completed. The utility of OptMAVEn is demonstrated through the design of broadly-neutralizing anti-influenza antibodies.