Folding Rate Prediction Using Total Contact Distance

Abstract

Linear regression analysis found that either contact order (CO) or long-range order (LRO) parameter has a significant correlation with the logarithms of folding rates. This suggests that sequence separation per contact and total number of contacts are both important in determining the rate of folding. Here, the two factors are incorporated into a new parameter, total contact distance (TCD). Using a database of 28 two-state or weakly three-state folding proteins, TCD is found to be the most accurate among the three parameters (CO, LRO, and TCD) in terms of correlation and prediction. It provides even more accurate prediction than the best neural network results with two descriptors (contact order and stability per residue). The improvement is achieved in all three-structural classes (all α, β, and mixed). The accuracy of total contact distance in predicting folding rates is essentially unchanged if “short”-ranged contacts ( |i − j| ≤ 14 ) are not included in calculation. Thus, only long-range contacts with a sequence separation of more than 14 residues are important in determining the rate of folding. This is consistent with the results from the long-range order parameter. One of the significant outliers in prediction is found to be associated with the only protein in the database that involves nonlocal disulfide bonds. Removing the protein leads to a correlation coefficient of 0.89 between experimental observed and predicted folding rates in jackknife cross validation. The corresponding values for CO and LRO are 0.71 and 0.80, respectively.