De Novo Protein Structure Prediction

An unparalleled amount of sequence data is being made available from large-scale genome sequencing efforts. The data provide a shortcut to the determination of the function of a gene of interest, as long as there is an existing sequenced gene with similar sequence and of known function. This has spurred structural genomic initiatives with the goal of determining as many protein folds as possible (Brenner and Levitt, 2000; Burley, 2000; Brenner, 2001; Heinemann et al., 2001). The purpose of this is twofold: First, the structure of a gene product can often lead to direct inference of its function. Second, since the function of a protein is dependent on its structure, direct comparison of the structures of gene products can be more sensitive than the comparison of sequences of genes for detecting homology. Presently, structural determination by crystallography and NMR techniques is still slow and expensive in terms of manpower and resources, despite attempts to automate the processes. Computer structure prediction algorithms, while not providing the accuracy of the traditional techniques, are extremely quick and inexpensive and can provide useful low-resolution data for structure comparisons (Bonneau and Baker, 2001). Given the immense number of structures which the structural genomic projects are attempting to solve, there would be a considerable gain even if the computer structure prediction approach were applicable to a subset of proteins. There are two approaches to predicting protein structure. Template-based methods identify one or more homologues on which the structure is based. Ab initio or de novo methods obtain a structure more directly from sequence, without the need for a template. De novo techniques are much more computationally intensive than template methods and are limited to smaller proteins (<100–150 residues). Templatebased methods can be applied to larger proteins and are generally more accurate than de novo methods. However, this is only true when a template exists [<2000 known folds in SCOP 1.61 (Murzin et al., 1995; Andreeva et al., 2004) out of an estimated 10,000 that are possible (Koonin et al., 2002)] and can be found and properly aligned. De novo methods are necessary when no template exists and competitive in accuracy when templates cannot be identified or aligned with confidence. Even when a good template and alignment are found, de novo methods are still necessary to build the nonhomologous “loop” regions. Perhaps more so than for other methodologies, the development of de novo methods has been greatly aided by the blind tests provided biannually by the Critical