The cavity method to protein design problem

In this study, we propose an analytic statistical mechanics approach to solve a fundamental problem in biological physics called protein design. Protein design is inverse of structure prediction, and its solution the amino acid sequence that best stabilizes given conformation. Despite recent rapid progress using deep learning, challenge exploring principles remains. Contrary previous computational studies, used cavity method, extension mean-field approximation becomes rigorous when interaction network tree. We found for small two-dimensional (2D) lattice hydrophobic-polar (HP) models, by method yields results almost equivalent those from Markov chain Monte Carlo with lower cost.