Characteristics of Protein Fold Space Exhibits Close Dependence on Domain Usage

Abstract

With the growth of the PDB and simultaneous slowing of the discovery of new protein folds, we may be able to answer the question of how discrete protein fold space is. Studies by Skolnick et al. (PNAS, 106, 15690, 2009) have concluded that it is in fact continuous. In the present work we extend our initial observation (PNAS, 106(51) E137, 2009) that this conclusion depends upon the resolution with which structures are considered, making the determination of what resolution is most useful of importance. We utilize graph theoretical approaches to investigate the connectedness of the protein structure universe, showing that the modularity of protein domain architecture is of fundamental importance for future improvements in structure matching, impacting our understanding of protein domain evolution and modification. We show that state-of-the-art structure superimposition algorithms are unable to distinguish between conformational and topological variation. This work is not only important for our understanding of the discreteness of protein fold space, but informs the more critical question of what precisely should be spatially aligned in structure superimposition. The metric-dependence is also investigated leading to the conclusion that fold usage in homology reduced datasets is very similar to usage across all of PDB and should not be ignored in large scale studies of protein structure similarity.