Simulation time in Umbrella sampling

GROMACS version: 2020.2
GROMACS modification: Yes/No

I am just beginning to learn MD (with GROMACS), so please excuse me if the questions are very basic. I am trying to find binding free energy of a complex between protein (xray crystal structure) and ligand (docked). Protein is approximately 200 residues and the ligand is approximately 800 Daltons. I am using Prof. Lemkul’s mdtutorials site and specifically the Umbrella Sampling protocol as available here.

I calculated the force and generated configurations using two reaction coordinates and got a graph like below.

I got 12 windows of 0.2nm width.

Now I have to do the actual sampling. Here is the problem!

On my desktop, at dt=2fs, 100 ps takes a little more than 3 hours. According to the instruction I should do 12 x 10ns calculations …which is equivalent to 150 days. I have another protein on which similar task needs to be done. Thus I am looking at ~300 days!! :-(
Sadly for me, I don’t have that much time (assuming my computer won’t die midway).

My question is:

  • Can I do 1ns simulation and yet compare the binding free energy of two proteins whose ligands are same? … or 1ns simulation is too less?

  • If I can’t which guideline or rule (established in the MD community) will be violated (if I choose 1ns simulation)?

  • In general, what is the guideline for the length of simulation time for the Umbrella sampling?

Any help will be really appreciated.

1 ns will work, but the accuracy will be less. sometimes there won’t be proper overlapping between the bins. It is recommended to have overlapping bins as mentioned in tutorials.
Try different binding energy calculations which required less computational power

As with any simulation, you have to prove that your results have converged and you can compute meaningful error bars. In 1 ns, I doubt your solvent will have equilibrated, particularly within the formerly occupied binding site. Other approaches may be more suitable if you can’t perform umbrella sampling, but any method that seeks to compute ΔG of binding is going to require considerable computational resources.

Thank you @scinikhil
I should explore other binding free energy calculations

Thank you @jalemkul

Your comment makes a lot of sense. Basically my computer is inadequate for Umbrella sampling calculations!! I should look for other alternatives like alchemical approach (thermodynamic cycle). The individual production runs seem to be only 1ns (at dt=2fs) and with a smaller unit cell (no need to pull), it seems I can finish the calculation in less time (1x30 of complex ~= 30ns and 1x20 of ligand ~=20ns).

One last question

Is ΔG calculations using the alchemical approach (as described here on, received in the community as a good alternative to the Umbrella Sampling?

alchemical approach is more computationally expensive and you need to give adequete time for the equilibration.