Affiliations: Centre de Biochimie Structurale, INSERM U 414, CNRS UMR 5048, Université de Montpellier I, Faculté de Pharmacie, 15, av. Ch. Flahault, F‐34060 Montpellier Cedex 2, France | Laboratoire de Biochimie Théorique, UPR 9080, IBPC, 11, rue P. et M. Curie, F‐75005 Paris, France
Abstract: We are here addressing the problem of the automatic determination of a protein structure at atomic resolution, by using only the signal recorded on three spectra: 2D 15N HSQC, 3D 15N NOESY‐HSQC and TOCSY‐HSQC. A modified version of the neural network RESCUE (J.L. Pons and M.A. Delsuc, J. Biomol. NMR 15 (1999), 15–26), N15‐RESCUE, is developed in order to predict the amino‐acid type from only the 15N, HN, Hα and Hβ chemical shifts. The spatial distances between protein residues are estimated by automatic comparison of columns extracted from a 3D 15N NOESY‐HSQC spectrum, using the FIRE method (T.E. Malliavin, P. Barthe and M.A. Delsuc, Theor. Chem. Accts 106 (2001), 91–97). The predictions provided by both FIRE and N15‐RESCUE methods are then used for the determination of a preliminary NMR structure of the protein p8. A mean RMSD value of 2.31±0.86 Å is observed between the coordinates of heavy atoms from helices αI and αII, and the αIII helix is taking random orientations with respect to the other helices. This random orientation is a consequence of the lack of predicted proximities between αIII and αII, and is in agreement with other independent observations made on p8 structure.
Keywords: Nuclear magnetic resonance, NMR, structure prediction, computer‐aided assignment, data processing, neural network, chemical shift index
Journal: Spectroscopy, vol. 17, no. 2-3, pp. 559-568, 2003
