This SBIR Phase I project will demonstrate the applicability of nonlinear recursive filtering techniques for the improved refinement of molecular structures from NMR data. These techniques evolved from well-known and widely used estimation methods used for precise navigation and control of aircraft and satellites. Through preliminary examples, the methodology compares favorably with existing NMR approaches and converges properly while maintaining reasonable computational costs. Additionally, the algorithms are robust to major gaps in measured data and generate a set of structures statistically compatible with the data. The algorithm takes particular advantage of the structure of the NMR data, replacing the high-order multiple-local-minima optimization process by a series of simple non-linear recursive steps. This is a powerful new tool that promises to improve the resolution of refined structures. Direct refinement through the NOE intensities becomes practical for very large molecules by a """"""""refinement wave"""""""" implementation that scales linearly with the number of structural parameters. In Phase I, the methods will be tested on large protein molecules (200 residues) and will be evaluated with various error models. In Phase II, the estimation algorithms will be modified as necessary and interfaced to a commercial structure determination code such as X-PLOR.
