Bayesian protein structure prediction and potentials of mean force: from hack to math

Thomas Hamelryck, University of Copenhagen. Part of the mathematics statistics seminars series.

Accurate and efficient prediction of the three-dimensional structure of proteins from their amino acid sequence is one of the great open problems in science. Many protein structure prediction methods make use of so-called “potentials of mean force” (PMFs). These PMFs are interpreted as physical energies whose parameters are estimated from the collection of known protein structures. However, for many decades the validity of these potentials has been widely disputed. I will explain how these potentials arise as the result of well-justified Bayesian reasoning. Specifically, PMFs result from the application of Jeffrey\'s rule or probability kinematics, which allows Bayesian updating in the light of updated information on the probabilities of the elements of a partition of the event space. This explanation validates PMFs and opens the way to a full Bayesian treatment of the protein structure prediction problem.

Thomas Hamelryck, University of Copenhagen