Accurate computational predictions of protein structure would be a very useful tool for understanding diseases related to (or caused by) proteins with unknown structure. Tertiary structure prediction algorithms based on a knowledge of secondary structure have been described for some time. A recently developed neural network method is able to predict secondary structure in one of three states with 75 percent accuracy. Furthermore, this method produces estimated probabilities of finding helix, sheet, or coil at every residue in a protein, which can be used to derive predicated ensembles of backbone dihedral angles. Using the quasichemical approximation, these predicted dihedral angle frequencies can be sued to create a pseudopotential which includes the effect of all sequentially local interactions on backbone dihedral angles. When combined with pseudopotentials representing non- local interactions, a useful energy function for protein folding simulations and structure prediction could be created. Potential applications of this model include more accurate secondary and tertiary structure prediction that can be obtained with current methods, and possible insight into the process of protein folding.
| Chandonia, John Marc; Cohen, Fred E (2003) New local potential useful for genome annotation and 3D modeling. J Mol Biol 332:835-50 |
