Algorithmic Methodologies for Discovery of Nonsequential Protein Structure Similarities
An increasing number of protein structures are becoming available that either have no known function or whose functional mechanism is unknown or incomplete. Using experimental methods alone to explore these proteins in order to determine their functional mechanism is unfeasible. For this reason, much research has been put into computational methods for predicting the function of proteins [5, 14, 31, 34, 44, 53]. One such computational method is functional inference by homology, where annotations from a protein with known function are transferred onto another protein on the basis of sequence or structural similarities.
Protein sequence comparisons have been used as a straightforward method for functional inheritance. If two proteins have a high level of sequence identity, frequently the two proteins have the same or related biological functions. This observation has been used as a basis for transferring annotations from a protein that is well characterized to a protein with unknown function when the two proteins have high sequence similarity [3, 4, 45]. Frequently, only the protein residues that are near the functional region of the protein are under evolutionary pressure for conservation. Therefore, the global sequence similarity may be relatively low while local regions within the two sequences maintain a higher level ...