Dynamics of proteins predicted by molecular dynamics simulations and analytical approaches: Application to α‐amylase inhibitor

The dynamics of α-amylase inhibitors has been investigated using molecular dynamics (MD) simulations and two analytical approaches, the Gaussian network model (GNM) and anisotropic network model (ANM). MD simulations use a full atomic approach with empirical force fields, while the analytical approaches are based on a coarse-grained single-site-per-residue model with a single-parameter harmonic potential between sufficiently close (r ≤ 7 Å) residue pairs. The major difference between the GNM and the ANM is that no directional preferences can be obtained in the GNM, all residue fluctuations being theoretically isotropic, while ANM does incorporate directional preferences. The dominant modes of motions are identified by (i) the singular value decomposition (SVD) of the MD trajectory matrices, and (ii) the similarity transformation of the Kirchhoff matrices of inter-residue contacts in the GNM or ANM. The mean-square fluctuations of individual residues and the cross-correlations between domain movements retain the same characteristics, in all approaches—although the dispersion of modes and detailed amplitudes of motion obtained in the ANM conform more closely with MD results. The major weakness of the analytical approaches appears, on the other hand, to be their inadequacy to account for the anharmonic motions or multimeric transitions driven by the slowest collective mode observed in MD. Such motions usually suffer, however, from MD sampling inefficiencies, and multiple independent runs should be tested before making conclusions about their validity and detailed mechanisms. Overall this study invites attention to (i) the robustness of the average properties (mean-square fluctuations, cross-correlations) controlled by the low frequency motions, which are invariably reproduced in all approaches, and (ii) the utility and efficiency of the ANM, the computational time cost of which is of the order of "minutes" (real time), as opposed to "days" for MD simulations. Proteins 2000;40:512–524. © 2000 Wiley-Liss, Inc.