Improved Physics-Based Single-Position Protein Sequence Redesign with a Residue-Pairwise Generalized Born Model

Computational protein design (CPD) aims to create proteins with new properties. Applications include the design of new catalytic reactions, new peptide ligands, vaccines, and new materials. A key element of CPD is the energy or scoring function used to discriminate the sequences and conformations. We use an energy function combining molecular mechanics (MM) with generalized Born (GB) solvation along with approximations that make the model pairwise decomposable. Our CPD approach is implemented in the Proteus software. The use of a physics-based energy function ensures a certain transferability and explanatory power to the model. An ambitious problem, often used to evaluate CPD approaches, is the redesign of full protein sequences in which the sequence of all positions is optimized at the same time. We obtained good results previously, with protein cores similar to natives and Superfamily recognition of the sequences close to 100%. A possibility to further improve our results consists of reducing the errors due to the pairwise decomposition of the solvation terms. Our group has proposed a "fluctuating dielectric boundary" (FDB) approach allowing an exact decomposition of the GB term. It was previously applied only to the sidechains. The goal of the present work is to extend the GB FDB approach to the whole protein and apply it to the single-position redesign of protein sequences. A notable improvement in the quality of designed sequences is obtained. This allows our Proteus program to have one of the most realistic electrostatic models among CPD approaches.