About the plotted result of umbrella sampling

User discussions
1

GROMACS version:2023.5
GROMACS modification: Yes/No
Here post your question
After upgrading my computer with the new GTX4070 ti graphic cards and several trials and errors and searching for problems others encoutered in umbrella sampling, I was finally able to finished one round of umbrella sampling with my protein-complex structure. It looked not that bad.

image
image1032×1338 112 KB

Here I have several questions:

  1. The top graph was generated from the hist.xvg file has a title of “Umbrella potential”. The curve that I remember from Justin’s tutorial is PMF or Pentential of Mean Force. I want to know if the title of the curve is correct or not?
  2. I have a another structure that is similar as the one I used for this umbrella sampling. If I do umbrella sampling with it, can I use less pulling distance as current one, because I found the curve in the top graph converge quite well after 4 nm of pulling distance?
  3. Do I need additional frames, such as those between 3 and 4 nm to make the curve smoother? Is current curve enough for the final result?
  4. I found one of the histograph (the green one) has two peaks, a larger one on the left and a small one on the right, that seems a bit of conterintuitive. Does anyone know what was going on?
    image
2

You’ve got very little overlap between many of the curves in the histogram, especially in the region 3.3 to 3.5, but not only there. I’d recommend adding more points/umbrellas.

Indeed, the green curve looks suspicious. I would suspect (but may be wrong) that the force constant is too low to keep a proper sampling distribution. If you have to increase the force constants, your histogram peaks will be more narrow and you will need even more umbrella states.

3

Hi MagnusL, thank you for your suggetions. I added additional frames and histogram overlap much better and the the initial curve become much smoother.

image
image1086×1504 140 KB

But here is another problem that I could not solve. Between histogram 7 and 8 (red arrow), which correspond to frame186 and 187, respectively the distance between the two frame is only 0.015 nm.
image
Should I do longer pulling time to increase the frames that sampling more frequently around these two frames? Should it help?
Also can I use different k value for different frame, as I don’t want to rerun the whole simulation with higher k value.
Update: I just found a previous post discussing this.

I will update this post when I got this area sampled!
Thank you for your help!