Item

Abstract

The principal difficulty in using nuclear magnetic resonance (NMR) data for biomolecular structure determination is not so much
experimental imperfections but approximate theories relating structure
to measurands. Furthermore, these theories are incomplete as they
involve auxiliary parameters which are not measurable. In order to
give a reliable picture of a biomolecule, structure determination
methods need to determine unknown parameters from definite rules and
ought to provide the uncertainty of the derived
coordinates. Conventional approaches neglect uncertainties of any kind
and therefore by definition fail to give an estimate of structural
reliability. In order to deal with auxiliary parameters, they resort
to heuristics which renders an objective interpretation of the
generated atom positions impossible. Recently, we have introduced a fully
probabilistic approach to structure determination from NMR data. We
describe here an extension of this approach which incorporates
inconsistent nuclear Overhauser effect and J-coupling
measurements. Auxiliary parameters are estimated along with the atomic
coordinates using Markov Chain Monte Carlo. We apply the method to
data sets for two small proteins.