Galaxy | Tool Preview

Alchemical Run (version 2019.1)
GRO input file with the merged structure.
TOP input file with the merged structure.
Number of steps for each free energy window.
Number of MD steps for NVT equilibration for each free energy window.
Number of MD steps for NPT equilibration for each free energy window.
Number of MD steps for production runs of each free energy window.
Seed to initialize random generator for random velocities.
Time step for integration.
Constraints may be required to keep the ligands with alchemical states in the active site.
Convert all bonds to constraints, or only those containing hydrogen atoms
Accuracy of LINCS algorithm. For normal MD simulations an order of 4 usually suffices, 6 is needed for large time-steps with virtual sites or BD.
Number of iterations to correct for rotational lengthening in LINCS. For normal runs a single step is sufficient.
Maximum angle that a bond can rotate before LINCS will complain / [deg]
Global scaling - values must be between 0 and 1.

What it does

When you need to run the simulation outside Galaxy, you can generate and down load all the input files from this tool.

  1. Simply download the output "Input files".
  2. Untar it using tar -xvf Galaxy3-[Input_files].tar
  3. Run the bash script inside ./gmx_fep.sh {number of FEP windows} - Give the number of FEP windows as an argument.

This will generate input .MDP files and run all the steps for all the FEP windows iteratively.


Rules of Thumb for Intermediate States (taken from alchemistry.org)

  • These rules are not the end-all set and you should be familiar with why each one is suggested before just accepting them.
  • Bonded terms can be modified/turned off linearly. This includes angle or bond force constants as well as unconstrained bond distances.
  • Constrained bonds should not change length. There are free energy changes that cannot be ignored affiliated with this action.
  • Maximize similarity between states by removing/decoupling as few atoms as possible.
  • Do not open and close rings. This supersedes the previous rule.
  • Statistical uncertainty between any neighboring states should be equal. Rather challenging to do, but it has been proven to have the lowest variance path if you can pull it off.
  • Deleting or adding atoms should always be done with a soft core potential.
  • Changes in parameters can be done linearly.
  • All charge on atoms must be turned off prior to atomic repulsion. Otherwise you can get an infinite attractive potential and crash your simulation.
  • Similarly for only changes in terms, it's generally more efficient to change electrostatic terms separate from Lennard-Jones terms.
  • More states is better than fewer. Variance shrinks rapidly with number of states. You want the difference between intermediaries to be between 2-3 kBT

Obviously you will be limited on CPU power. Fewer states also leads to more samples begin required from each state, so take this into account when deciding number of states as well. However, for MBAR and TI, it can be shown that spreading samples across multiple states does not significantly affect the uncertainty, since for TI, each state contributes less to the total uncertainty, and in MBAR, data contributes to the statistical precision of states with similar values of lambda.

Shape of the variance does not significantly change with number of atoms, only magnitude. More intermediates will still be required for a large number of atoms to reduce statistical noise. Charge should be maintained across all λ

Simply having charged molecules is fine, but the net of the system should remain constant. If you must change the net charge, there are complicated ways to do so.

Short prototype simulations are recommended. Even as short as 100 ps, the prototypes can provide rough magnitude of variance estimates, although will likely under-predict the free energy as many configurations remain unsampled.


Input

  • .GRO input
  • .TOP input

Output

  • TI/FEP data
  • TI/FEP trajectory
  • Report