1: \begin{abstract}

2:   We describe different Bayesian ensemble refinement methods, examine

3:   their interrelation, and discuss their practical application.  With

4:   ensemble refinement, the properties of dynamic and partially

5:   disordered (bio)molecular structures can be characterized by

6:   integrating a wide range of experimental data, including

7:   measurements of ensemble-averaged observables.  We start from a

8:   Bayesian formulation in which the posterior is a functional that

9:   ranks different configuration space distributions.  By maximizing

10:   this posterior, we derive an optimal Bayesian ensemble

11:   distribution. For discrete configurations, this optimal distribution

12:   is identical to that obtained by the maximum entropy ``ensemble

13:   refinement of SAXS'' (EROS) formulation.  Bayesian replica ensemble

14:   refinement enhances the sampling of relevant configurations by

15:   imposing restraints on averages of observables in coupled replica

16:   molecular dynamics simulations.  We show that the strength of the

17:   restraint should scale linearly with the number of replicas to ensure

18:   convergence to the optimal Bayesian result in the limit of infinitely many

19:   replicas.  In the

20:   ``Bayesian inference of ensembles'' (BioEn) method, we combine the

21:   replica and EROS approaches to accelerate the convergence.  An

22:   adaptive algorithm can be used to sample directly from the optimal

23:   ensemble, without replicas. We discuss the incorporation of

24:   single-molecule measurements and dynamic observables such as

25:   relaxation parameters.  The theoretical analysis of different

26:   Bayesian ensemble refinement approaches provides a basis for

27:   practical applications and a starting point for further

28:   investigations.

29: \end{abstract}

30: