A0363
Title: Maximum entropy ensemble refinement
Authors: Benjamin Eltzner - Max Planck Institute for Multidisciplinary Sciences (Germany) [presenting]
Bert de Groot - Max Planck Institute for Multidisciplinary Sciences (Germany)
Michael Habeck - University of Jena (Germany)
Daniel Rudolf - University of Passau (Germany)
Julian Hofstadler - University of Passau (Germany)
Abstract: In some cases, features measured from a protein ensemble, like atom distances, are not recovered on average in molecular dynamics simulations. The problem is approached from the maximum entropy point of view. The problem then presents as a Bayesian inference problem with unknown likelihood normalization, a so-called doubly intractable problem. This type of problem requires sophisticated two-step Monte Carlo methods. The focus is on NOE measurements, where the measured peak intensities are proportional to the inverse sixth power of atomic distance. Ensembles derived using the corresponding maximum entropy energy terms in a modified molecular dynamics simulation show a wider variety of structures than ensembles derived by other refinement methods while still satisfying measured feature bounds on average.
