Test System Repository

With the recent proliferation of free energy methods, it seems that it would be beneficial to have a more standardized test set of free energies.

One of the biggest problems of fine comparisons of methods is the difficulty of trying to agree on a single number for the free energy. If one is not sure of the value of the free energy, fine comparisons of methods are very difficult. Additionally, different programs with different bookkeeping, or parameters that have been rounded in some way can have legitimate small differences in the free energies, obscuring differences in the methods. Therefore, except for the simplest system mentioned here (UA methane in water), these are all zero total free energy.

With a zero transformation, then it is necessary to partially specify the path, so that particularly clever methods do not manage to solve the problem in ways that would not be valid in general simulations. Because of this, the zero transformations here are defined with the endpoints and one midpoint along the transformation. This ensures that a large transformation is performed, but allows these systems to be used to improve free energy pathways as well.

Problem 1) Is the method at all valid for molecular systems?

   * System: Simplest molecular free energy system = UA methane in TIP3P water.

   * Notes: There are no bond, angle or torsions terms, or solute/solvent charge:this is the simplest system that can be truly defined as realistic.

Problem 2) Can the free energy method handle multiple atomic sites efficiently?

   * System: Napthalene null transform with benzene as intermediate, in TIP3P water

   * Notes: No ligand degrees of freedom to complicate the analysis. If this is too easy, could also perform the anthracine->benzene->anthracine transformation.

Problem 3) Can the method handle water rearrangement around charges?

   * System: two LJ spheres, tethered together, with +1/-1 charges, with reversal the charges.

   * Notes: This setup allows avoidance of computing free energies of ions directly, which is still not handled completely correctly in many codes.

Problem 4) Can the method handle torsional degrees of freedom?

   * System: 1-octanol -> ethane -> 1-octanol in TIP3P water.

   * Notes: Topologically, the system would be set up as HO-(CH2)_14-OH, with the middle two carbons remaining coupled to the system for the entire transformation. The h-bonds between alcohols and water might hopefully slow down the torsional sampling).

Problem 5) Can the method handle complications caused by putting all together in complex systems?

   * System: Complicated substituted aromatic like Imatinib, with three substituted positions, with the transformation to cycle the substituents to different positions along the aromatic with benzene as the intermediate.

Remaining questions for setup:

   * Size of system? As small as we can make it for each system. Provide 10 A from the edge of interacting molecules.

   * Simulation parameters: what electrostatic and other cutoff parameter should be set? What temperature and pressure control methods should be used?

Estimators of the uncertainty should be validated against uncertainty generated directly from runs from independent configurations (at least 40), and should include the computation of the correlation time of the observable used to calculate the free energies (such as the potential energy differences or dH/dL).

1. Input topology and parameter files in a number of different formats:

   * GROMOS (400 Euro)

   * CHARMM ($600)

   * GROMACS (we have)

   * AMBER (we have)

   * NAMD (free)

   * DESMOND (we have)

   * DL_POLY (free)

   * TINKER (free)

   * LAMMPS (free)

2) Independent prequilibrated starting configurations for each system (40-100). We will specify initial box size, velocities, positions.

3) Exact energies of the starting configurations to make it easy to verify input files for additional programs.

4) Results from a number of different methods (TI, BAR, WHAM, Wang-Landau recursion).

   * TI: from Gromacs 4.0

   * BAR, EXP, MBAR:

   * Wang-Landau:

   * Transition Matrix approaches 

5) Optimization of variables

5a) Equilibrium methods

   * For all: Spacing of states, pathway

   * TI: no others

   * BAR: no others

   * MBAR:no others

   * DCMBAR: size of blocks, approximations in the dimension reduction of control variates.

5b) Equilibrium-at-limit methods

   * For all : Spacing of states, pathway, MC or Gibbs sampling step type used

   * Wang-Landau: The degree of flatness for decrementing the weight step, the magnitude of the weight step

   * Transition state approaches: The transition kernel used