Sequential Monte Carlo Methods for Protein Folding

摘要: We describe a class of growth algorithms for finding low energy states heteropolymers. These polymers form toy models proteins, and the hope is that similar methods will ultimately be useful native real proteins from heuristic or priori determined force fields. share with standard Markov chain Monte Carlo they generate Gibbs-Boltzmann distributions, but are not based on strategy this distribution obtained as stationary state suitably constructed chain. Rather, growing polymer by successively adding individual particles, guiding towards configurations lower energies, using "population control" to eliminate bad increase number "good ones". This done via breadth-first implementation in genetic algorithms, depth-first recursive backtracking. As seen various benchmark tests, resulting extremely efficient lattice models, still competitive other simple off-lattice models.