Progress in super long loop prediction.

摘要: Sampling errors are very common in super long loop (referring here to loops that have more than thirteen residues) prediction, simply because the sampling space is vast. We developed a dipeptide segment algorithm solve this problem. As first step evaluating performance of algorithm, it was applied problem reconstructing native protein structures. With newly constructed test set 89 ranging from 14 17 residues, method obtains average/median global backbone root-mean-square deviations (RMSDs) structure (superimposing body protein, not itself) 1.46/0.68 A. Specifically, results for various lengths 1.19/0.67 A 36 fourteen-residue loops, 1.55/0.75 30 fifteen-residue 1.43/0.80 sixteen-residue and 2.30/1.92 9 seventeen-residue loops. In vast majority cases, locates energy minima lower or equal minimized loop, thus indicating new successful rarely limits prediction accuracy. Median RMSDs substantially averages small number outliers. The causes these failures examined some detail, can be attributed flaws function, such as pi-pi interactions accurately accounted by OPLS-AA force field we employed study. By introducing model which has superior description interactions, significantly better were achieved quite few former Crystal packing explicitly included order provide fair comparison with crystal