A combined approach for protein structure prediction
EMSL Project ID
8491a
Abstract
The protein structure prediction is one of the grand challenges in modern science. The problem consists of determining the tertiary structure of a protein given its primary sequence of amino acids. Experimental approaches, like X-ray crystallography and NMR spectroscopy, are very time consuming. Computer simulations that predict the protein fold are a promising alternative because they not only tackle the folding of existing proteins but also the folding of engineered proteins, including those designed for drug purposes. Two kinds of methods have been identified to tackle the protein structure prediction problem: knowledge-based methods that rely on the presence of homologous proteins (in sequence or structure) in the databases, and physics-based methods that emphasize more the physical principles. Although the physics-based methods are important to find genuine new folds, they are also extremely expensive. Thus, knowledge-based methods are a valid alternative when there is some homology. However, because in most cases only fragments of the proteins are similar, the results achieved by these methods may be limited. We have developed a new methodology for protein structure prediction that uses results from known proteins in combination with a physics-based method. This methodology is composed of two phases: the first phase generates initial configurations using secondary structure prediction as well as fold recognition servers. The second phase improves those initial configurations through a sophisticated global minimization algorithm that treats the full-dimensional global optimization problem as a series of small-dimensional ones. The minimization space is the set of dihedral angles. Because the initial configurations already have some structure formed, this method optimizes the space of dihedral angles for which not much information from existing proteins was found. Thus, the method works on subspaces of the original space of dihedral angles.
To support this methodology we have developed ProteinShop, an interactive visualization tool for protein modeling with the goal of manipulating protein structures with pinpoint control, guided in large part by the user's biological and experimental instinct. ProteinShop takes a given sequence of amino acids and uses visualization guides to help generate secondary structures according to predictions, identifying alpha-helices and beta-strands and the coil regions that bind them. Once secondary structures are in place, researchers can twist and turn these configurations until they come up with a number of possible tertiary structures conformations.
ProteinShop uses concepts from robotics and animation which allow researchers to re-configure structures without breaking them. The use of inverse kinematics enables all the angles in the structure to move as jointed segments simulating the way joints in our bodies interrelate. Whereas it used to take days of computational time to create initial configurations, this process can now be accomplished in a few hours. In addition, ProteinShop allows us to experiment among various possible conformations while simultaneously monitoring the energy profile of the protein.
Furthermore, ProteinShop allows us to manipulate structures generated by the global optimization process as it runs and then put them back in the queue of possible configurations for further optimization with the goal of dynamically steering the search through the vast conformational space. This feature permits to add human knowledge and intuition to the protein structure prediction process, thus bypassing those bad configurations that would otherwise be fruitless for optimization. This approach saves compute cycles and accelerates the entire process; therefore, more and larger problems can be attempted.
Project Details
Project type
Exploratory Research
Start Date
2005-01-01
End Date
2006-01-18
Status
Closed
Released Data Link
Team
Principal Investigator
Team Members