(467j) ? Dynamics: An Emerging Method for Physics-Based Protein Design

Protein design is useful in many fields including medicine, biocatalysis, and biotechnology. Computational protein design is undergoing a revolution as deep learning approaches displace physics-based approaches like Rosetta that have historically suffered from low success rates. In spite of this, a need remains for more accurate physics-based protein design methods to complement and inform deep learning approaches in situations where training data is sparse, such as designing molecular machines or utilizing non-natural amino acids. Since many protein design methods can be reduced to free energy optimization problems, alchemical methods, which estimate free energy from molecular simulation, provide a more rigorous and accurate alternative to existing methods like Rosetta. While most alchemical methods are too expensive to be practical for protein design, an emerging alchemical method known as λ dynamics is uniquely suited to efficiently explore the large sequence spaces encountered in protein design. Retrospective and prospective studies with λ dynamics have demonstrated its high accuracy. New methods are enabling λ dynamics to search even larger numbers of sequences, and the utility of these methods in screening for favorable mutations in protein G will be described. Ongoing applications of λ dynamics in protein design include antibody design benchmarks, enzyme redesign, and conducting peptide nanowire design.