Book of Abstracts: Albany 2011
June 14-18 2011
©Adenine Press (2010)
Protein flexibility methods to compare protein structure predictions
Geometric measurements such as RMSD or GDT score are routinely used as a preliminary tool to assess the quality of a model to a certain reference structure (usually high resolution X-ray or NMR data). Two problems with these measures are: (1) at a given temperature there is not one single structure, but an ensemble of them; (2) geometric analysis gives global information, is highly degenerate and does not take the protein’s topology into consideration. We investigating the utility of using an energetic measure based on a protein’s intrinsic flexibility pattern to complement geometry measurements in model quality evaluation. To account for flexibility we use the elastic network model (ENM) theory [1-3], which is computationally inexpensive and gives an accurate approximation to protein flexibility. In ENM the protein is represented as a series of nodes (Calphas). A simple Hamiltonian is derived by: (1) finding all pairs of nodes closer than some cutoff distance; and (2) connecting them via springs. Calculation of the hessian and its diagonalization yields a set of eigenvector-eigenvalue pairs representing the directions of maximum deformability in the protein system. This information can be used to calculate deformation energies needed to deform the reference structure into the model one. This energy will rapidly increase when deforming regions of the protein that are not intrinsically flexible. Thus, two structures with the same RMSD (or same GDT score) might have very different energies. Conversely, two structures that have the same energy value might be far apart in RMSD (see figure below). Having an energetic measure is helpful in identifying structures within thermal noise of each other (e.g. within the native ensemble), meaning that the model is of good enough quality. It might be useful in events such as CASP  to quantify how hard refinement of a given template is going to be.
Figure: Top is a reference crystal structure in CASP9 (TR569). Below are two submitted models. The one on the left would receive a higher score in geometric measurements, but the main difference is in a very flexible area of the protein, resulting in similar deformation energies for both models
AP acknowledges support from EMBO long-term fellowship.
Laufer Center for Physical and Quantitative Biology, Stony Brook University, Department of Pharmaceutical Chemistry, University of California San Francisco, and University of Pittsburgh