UCLA

We develop and implement a brand new method for protein design. Our design method utilizes advances in protein energy scoring, Markov state modeling, machine learning, distributed computing, and optimization. First we generate a Markov State Model for human ubiquitin. This creates 100 separate fixed backbone conformations of the protein. Next, we generate quick and accurate energy functions for each of the 100 separate conformations utilizing the Rosetta energy function, mean field energy modeling, and stochastic gradient descent. Finally, we use these coarse-grained sequence-energy functions to design mutant sequences with specified conformational dynamics through use of optimization algorithms. We selectively stabilize target states through maximizing Boltzmann distribution probability for either one state or a pair of states. We showcase that we can accurately design sequences with altered energy landscapes for ubiquitin mutants up to eight mutations away from wild type ubiquitin using our design approach.