Some fundamental aspects of building protein structures from fragment libraries

Abstract

We have investigated some of the basic principles that influence generation of protein structures using a fragment-based, random insertion method. We tested buildup methods and fragment library quality for accuracy in constructing a set of known structures. The parameters most influential in the construction procedure are bond and torsion angles with minor inaccuracies in bond angles alone causing >6 A CalphaRMSD for a 150-residue protein. Idealization to a standard set of values corrects this problem, but changes the torsion angles and does not work for every structure. Alternatively, we found using Cartesian coordinates instead of torsion angles did not reduce performance and can potentially increase speed and accuracy. Under conditions simulating ab initio structure prediction, fragment library quality can be suboptimal and still produce near-native structures. Using various clustering criteria, we created a number of libraries and used them to predict a set of native structures based on nonnative fragments. Local CalphaRMSD fit of fragments, library size, and takeoff/landing angle criteria weakly influence the accuracy of the models. Based on a fragment's minimal perturbation upon insertion into a known structure, a seminative fragment library was created that produced more accurate structures with fragments that were less similar to native fragments than the other sets. These results suggest that fragments need only contain native-like subsections, which when correctly overlapped, can recreate a native-like model. For fragment-based, random insertion methods used in protein structure prediction and design, our findings help to define the parameters this method needs to generate near-native structures.

Figure 1.

Testing parameters used to build models with torsion angles. Both box-and-whisker plots indicate the effect on the global CαRMSD (between the original structure and reconstructed model) of specifying precise bond angles and bond lengths instead of using ideal values (Engh and Huber 1991). The R statistical computing environment was used to create all box-and-whisker plots (http://www.rproject.org/). The whiskers are set to either 1.5 the interquartile length or the most extreme data point if it is less. (A) Distribution of the CαRMSD for 1894 native protein structures rebuilt using increasing amounts of native information. The length dependence of the reconstruction routine is shown in the inset. In all of the box plots, we saw longer proteins near the high extremes of CαRMSD and smaller proteins near the lower extremes. (B) The same plot as the previous, except it was created using 1894 idealized protein structures.

Figure 2.

A β-strand (EEE EEE EEE) fragment cluster. This is one cluster from the all β-strand supercluster. (A) The 9mers are shown superimposed. It is in this position that a CαRMSD between two 9mers is calculated and clustering is executed. (B) The cluster is shown when the first residue of each 9mer is artificially superimposed. The latter more closely represents the effect of inserting 9mers into a protein structure.

Figure 3.

Testing the accuracy of fragment libraries. We analyzed the impact of the size of our fragment library as well as the method used to select the best replacement fragment. (A) The relationship between protein size and the global CαRMSD is shown. (B) The relationship between local CαRMSD (9mer to 9mer) and global CαRMSD is shown.

Figure 4.

Effect of takeoff angles on building protein models. The initial ψ torsion angle of each 9mer has an effect on any protein into which it is inserted when compared to the same fragment with a different takeoff angle. Eight 9mers with identical main chain atoms but differ by their initial ψ torsion angle. The α-carbons all perfectly overlay (large spheres), so the CαRMSD between these would be 0 Å. The figure was created using the Spock molecular graphics program (http://quorum.tamu.edu/spock/).

Figure 5.

Testing of the native insert algorithm. The native insert 9mer selection algorithm (where 161 proteins from the PISCES set were used) is compared to the CαRMSD selection algorithm. The numbers under the X-axis labels indicates the average local CαRMSD calculated from the 161 proteins for comparison. The global CαRMSD is calculated between the native coordinates and the model constructed from the selected 9mers. Interestingly, a lower local CαRMSD has a higher global CαRMSD.

Figure 6.

Global CαRMSD differences between idealized and native data structures. For the Kolodny structure set, the CαRMSD of the idealized structure to the native is plotted against number of residues.