A protein folding potential that places the native states of a large number of proteins near a local minimum

摘要: We present a simple method to train potential function for the protein folding problem which, even though trained using small number of proteins, is able place significantly large native conformations near local minimum. The training relies on generating decoys by energy minimization current and physically meaningful objective (derivative with respect torsion angles at conformation) during quadratic programming conformation also compare performance three different types functions find that while pairwise trainable, solvation itself untrainable if are generated minimizing starting conformation. best results obtained when interaction used function. six proteins which places total 42 within ~4 A rmsd 71 ~6 minimum out 91 proteins. Furthermore, threading test same ranks 89 be first other two as second.