Protein design for diversity of sequences and conformations using dead-end elimination

Proteins, especially antibodies, are widely used as therapeutic and diagnostic agents. Computational protein design is a powerful tool for improving the affinity and stability of these molecules. We describe a protein design method which employs the dead-end elimination (DEE) and A* discrete search algorithms with a few improvements aimed at making the procedure more useful for actual projects to design proteins for better affinity or stability. DEE/A* and related algorithms allow vast search spaces of protein sequences and their alternative side chain conformations ("rotamers") to be systematically explored, to find those with the best free energy of folding or binding. To maximize a protein design project's chance of success, it needs to find a diverse set of sequences to experimentally synthesize. It should also find structures that score well, not only on the pairwise-additive energy function which DEE/A* and related search algorithms must use, but also on a post-search energy function with accurate treatment of solvation effects. Straight DEE/A*, however, typically finds vast numbers of very similar low-energy conformations, making it infeasible to find a diverse set of sequences or conformations. Herein, we describe a three-level DEE/A* procedure that uses DEE/A* at the level of sequences, at the level of rotamers, and at an intermediate "fleximer" level, to ensure a wide variety of sequences as well as a diverse set of conformations for each sequence.A physics-based method is also described herein for calculating the free energy of folding based on a thermodynamic cycle with a model of the unfolded state. The free energies of both folding and binding may be used for the final evaluation of the designed structures. For example, when designing for improved affinity (binding), we can also ensure that stability is not degraded by screening on the free energy of folding.