Progress in super long loop prediction

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