Biomolecular sampling

Here are some thoughts for biomolecular sampling:

Different issues:

• What are the problems that are being solved?
• What are the different theoretical approaches/philosophies?
• What are the domains of applicability?
• What can be done now?
• What are the challenges?
• Examples of successes?
• Are there broad classes of algorithm that can be used to organize our thinking/writing? Here is a first stab at this kind of organization:
  • Biased dynamics algorithms, dating at least to Voter's PRL, and continuing through Mcammon's "accelerated MD" and then to Metadynamics
    • A critical distinction needs to be made between methods that bias the potential energy surface, and those that bias a PMF
    • The discussion of metadynamics and other PMF bias methods obviously links in to the discussion of CVs.
  • Algorithms which attempt to sample the hard to sample by mixing in states from an easier to sample distribution, dating at least to the replica exchange approach of Swendsen.
    • These are subdivided into exchange-type methods (multiple parallel walkers) and expanded-ensemble methods (single walker methods)
  • Umbrella sampling, which seems in a way to bridge these two: It can be viewed as sampling on a biased potential energy surface, or as mixing together samples under different potentials.
  • Nondynamical methods
    • Library Monte Carlo

Some topics:

• Protein folding
  • Multicanonical
  • Umbrella sampling
    • Torrie and Valleau The original paper on umbrella sampling
    • Other papers
  • Metadynamics
• Sidechain motion

• Rigid body motion

General Issues

• Defining collective variables
• Sampling vs Search: This is an important distinction that is often glossed over in review articles
  • Sampling algorithms asymptotically converge to a known distribution (usually Boltzmann)
  • Search algorithms do not guarantee any particular distribution, but can be useful as configuration generation engines
  • toppapers.org