NMR structure computation

Reduced representations are useful for determining the protein fold from limited or approximate NOE data. They are also useful as a step toward computing the all-atom structure. Because they involve fewer distinct restraints, and because the computational cost of molecular dynamics simulation or distance geometry algorithms increase as the square of the number of particles or faster, computation of the fold using reduced representations is extremely rapid. We are exploring the application of reduced representations to problems such as homology modeling and NMR spectrum assignment, in addition to determination of overall protein fold.

An important problem in NMR structure determination is determining the relative weights to be applied to restraining potentials for different NMR parameters (such as NOEs, coupling constants, and chemical shifts). This is closely related to the problem of defining "accuracy" in the context of NMR structure determination: there is no consensus about how to quantify the level of agreement between structural models and experimental data. We are developing a Bayesian formalism that specifies how to determine the weight and form of the restraint potentials, putting the different types of experimental constraints on equal footing.

Aims: We are developing a new computer program, MoSART (Molecular Structure Analysis and Refinement Tool) that will incorporate Bayesian restraint potentials for refining biomolecule structures based on empirical NMR constraints, including NOEs, vicinal and residual dipole coupling constants, and spin label induced perturbations to chemical shifts and relaxation rates. MoSART will implement the Amber force field for all-atom structure refinements and our "2p" force field for low resolution refinement, as well as tools for converting between all-atom and reduced representations. MoSART will be capable of refining ensembles of structures as well as individual structures against empirical constraints. An important component of MoSART will be tools for analyzing the accuracy and precision of structures. One novel tool will enable three-dimensional visualization of the distribution of residual strain in structures. We plan to distribute MoSART using an open source model in order to avoid the restrictions (and costs) associated with current structure calculation software packages.