LOCAL INDICATORS OF MODEL QUALITY Zhou G., Blanc, E. & Chapman, M. S.; Institute of Molecular Biophysics, Florida State University, Tallahassee, FL 32306-3015, USA.

Reciprocal space R-factors are a measure of overall model quality that are not sensitive to local detail. Jones et al. [Acta Cryst., (1991), A47, 110-119] suggested the use of real space R-factors, calculated from a local comparison of observed and calculated electron density near individual residues, to detect severe errors such as frame shift of sequence assignment. We use random shifts while keeping good geometry to test the sensitivity of local R-factors to see whether more subtle errors can be detected. Previous random shift tests did not keep good geometry. Unfortunately, this changes the overlap of the electron density of covalently linked atoms and has substantial impact on the real space R-factors. This is unlikely to occur in actual protein structure determinations with the imposition of geometrical restrains. Detection of error is therefore more difficult than suggested by these previous tests. In the current work, tests with random shift models, in which good geometry is maintained, have shown that real space R-factors are correlated to introduced error, but weakly. The R-factors of the most incorrect regions rise above noise when averaged over several residues. We will present the results of systematic tests with maps and models of various qualities at various resolutions.