Bayesian estimation of NMR restraint potential and weight: A validation on a representative set of protein structures

Authors

  • Aymeric Bernard,

    1. Unité de Bioinformatique Structurale, CNRS URA 2185, Institut Pasteur, 25-28 rue du Dr. Roux, Paris 75724, France
    Search for more papers by this author
  • Wim F. Vranken,

    1. Protein Data Bank in Europe, European Bioinformatics Institute, Wellcome Trust, Genome Campus, Hinxton, Cambridge, United Kingdom
    Current affiliation:
    1. Structural Biology Brussels, Vrije Universiteit Brussel, Pleinlaan 2, Brussel 1050, Belgium
    Search for more papers by this author
  • Benjamin Bardiaux,

    1. Leibniz-Institut für Molekulare Pharmakologie Campus Berlin-Buch, Robert-Roessle-Str. 10, Berlin DE 13125, Germany
    Search for more papers by this author
  • Michael Nilges,

    1. Unité de Bioinformatique Structurale, CNRS URA 2185, Institut Pasteur, 25-28 rue du Dr. Roux, Paris 75724, France
    Search for more papers by this author
  • Thérèse E. Malliavin

    Corresponding author
    1. Unité de Bioinformatique Structurale, CNRS URA 2185, Institut Pasteur, 25-28 rue du Dr. Roux, Paris 75724, France
    • Unité de Bioinformatique Structurale, CNRS URA 2185, Institut Pasteur, 25-28 rue du Dr. Roux, Paris 75724, France
    Search for more papers by this author

Abstract

The classical procedure for nuclear magnetic resonance structure calculation allocates empirical distance ranges and uses historical values for weighting factors. However, Bayesian analysis suggests that there are more optimal choices for potential shape (bounds-free log-harmonic shape) and restraints weights. We compare the classical protocol with the Bayesian approach for more than 300 protein structures. We analyze the conformation similarity to the corresponding X-ray crystal structure, the distribution of the conformations around their average, and independent validation criteria. On average, the log-harmonic potential reduces the difference to the X-ray crystal structure. If the log-harmonic potential is used, the constant weighting tightens the distribution around the average conformation, with respect to the distributions obtained with Bayesian weighting. Conversely, the structure quality is improved by the Bayesian weighting over the classical procedure, whereas constant weighting worsens some criteria. The quality improvement obtained with the log-harmonic potential coupled to Bayesian weighting validates this approach on a representative set of protein structures. Proteins 2011. © 2011 Wiley-Liss, Inc.

Ancillary