Dear Tomasz,

There is no universally accepted protocol. In my opinion the only emerging consensus is that the best results you get with a carefully prepared system. It means before you do an initial minimization of experimentally derived structure you should relieve all the offending conformations by hand. There are scattered pieces of software that might help. I personally like the best software pieces by Dokholyan. After minimization you should gradually raise the temperature from around 0 to around the final temperature you want to run your simulation. Taking into account all the caveats I mentioned in my previous post how you raise the temperature (coupling to solvent of all atoms do not matter that much but the speed with which you do it matters). As Nubia noticed you can follow the changes in variables by displaying them graphically. The use of NVT protocol for this first stage does not make much sense so use a standard NPT. I personally preferred coupling to solvent alone for a very simple reason that coupling to all atoms introduces artifacts such as local overheating that later needs to be released while heating of solvent alone is much better described from a thermodynamic point of view. In reality what matters the most is not that you follow a specified protocol as some of this discussion would like to believe but to create your own intuition by running several times similar simulations. Inherent chaos in MD makes it difficult to compare them but if you are in a similar dynamic regime they should be similar.

In reality there is a simple test for "equilibration" i.e. you could compare several short excerpts form a single long time simulation with each other and the appropriate time and space averages. This general remark has some weaknesses though. Proteins by definition are metastable systems (i.e. every single one has a specific time horizon of existence) so the test would have to be appropriately gauged for the particular system and most likely the details of the simulation like: how big a box is, how many counter ions, method of temperature control, a time progression algorithm, time step size, solvent treatment, electrostatics approximations etc.) would destroy this info without a careful study of the particular system.

Dear Tomasz,

You have asked a fascinating question that has many, if not more than 50 shades of gray.

On the surface it is a simple, straightforward question asking when, from a practical point of view, you can use your MD run to study a molecular detail you are interested in.

Depending on the experience, the origin of the group and problems studied you will get different answers ranging from several tens of picoseconds to 5 nanoseconds (all published data).

The real crux of the matter is that there is no solid theory connecting physical dynamics with thermodynamics. This connection is weak and tenuous and the theory of ergodicity became a separate field of math. The matter is farther complicated by a very approximate nature of the MD force fields and existence of chaos in MD simulations. Most of these problems are carefully swept under the carpet of popular applications.

An additional difficulty comes with an intended use of the simulation. For a purpose of checking how stable the folded architecture is, even 5-10 picoseconds would be sufficient, as the longest simulation done does not extend beyond microseconds. (The longest and very approximate simulations of folding extended to milliseconds but only real millionaires can afford that and believe in the results.)

However for the purposes of folding, or binding small ligands, or protein protein interactions even milliseconds would not "equilibrate" your system.

Even a meaning of equilibration is not clear in such cases as most of these events are not equilibrium effects.

So the advice is simple: Read the relevant literature and apply the limits cited in the type of paper you want to write. As a final advice do not question the real science behind it because then it becomes almost impossible to publish . As a corollary I have published a short paper on equilibration in Molecular Simulations proposing my own solution to the problem. Good Luck.

Dear Tomasz,

I do a graphical analysis to verify whether the system is balanced enough to run molecular dynamics simulation (like pressure, density, temperature...)

You could follow the tutorials of GROMACS package to do the equilibration before running MD.

Thank for Your answers. But I'd like to know which strategy for protein equilibration is the best for globular protein? I saw many modification this step, for exemple minimization only solvent and then minimization all system, equilibration protein with constraints or heating, equilibration only in NPT or first in NVT and than NTP. Is there some way to decide which steps would be best to equilibrate my system?

Dear Tomasz,

There is no universally accepted protocol. In my opinion the only emerging consensus is that the best results you get with a carefully prepared system. It means before you do an initial minimization of experimentally derived structure you should relieve all the offending conformations by hand. There are scattered pieces of software that might help. I personally like the best software pieces by Dokholyan. After minimization you should gradually raise the temperature from around 0 to around the final temperature you want to run your simulation. Taking into account all the caveats I mentioned in my previous post how you raise the temperature (coupling to solvent of all atoms do not matter that much but the speed with which you do it matters). As Nubia noticed you can follow the changes in variables by displaying them graphically. The use of NVT protocol for this first stage does not make much sense so use a standard NPT. I personally preferred coupling to solvent alone for a very simple reason that coupling to all atoms introduces artifacts such as local overheating that later needs to be released while heating of solvent alone is much better described from a thermodynamic point of view. In reality what matters the most is not that you follow a specified protocol as some of this discussion would like to believe but to create your own intuition by running several times similar simulations. Inherent chaos in MD makes it difficult to compare them but if you are in a similar dynamic regime they should be similar.

In reality there is a simple test for "equilibration" i.e. you could compare several short excerpts form a single long time simulation with each other and the appropriate time and space averages. This general remark has some weaknesses though. Proteins by definition are metastable systems (i.e. every single one has a specific time horizon of existence) so the test would have to be appropriately gauged for the particular system and most likely the details of the simulation like: how big a box is, how many counter ions, method of temperature control, a time progression algorithm, time step size, solvent treatment, electrostatics approximations etc.) would destroy this info without a careful study of the particular system.