I'm currently running a protein simulation on GROMACS. I was originally using CHARMM22/CMAP for the simulation, but with some reasons I decided to switch to CHARMM36.
The thing is, I expected CHARMM36 would not be that different from CHARMM22/CMAP (or at least will give a better result), but actually CAHRMM36 based simulation is giving me much higher RMSD value (~0.5 nm) compared to CHARMM22/CMAP (0.3~0.4 nm) based one. Also related to this, the output .gro file shows the bigger structural change of the protein after the 200 ns production run compared to the original structure.
My questions are: What could be the possible reasons for this? Does this mean CHARMM36 is inferior to CHARMM22/CMAP for my particular protein simulation?
Since I read the improvement of CHARMM36 from CHARMM22/CMAP in several papers, I am a little confused with this result.
Just to give more details of my system:
- The protein size = about 45,000 atoms
- System = water box, no charge
- em, t-coupling (around 300K) and p-coupling (1atm) all done and all converged
- The simulation runs under NPT ensemble for 200 ns, the RMSD gets stable around 100ns for both CHARMM22/CMAP and CHARMM36 cases
Thank you!
Wojciech Kopec
Thank you so much for your help.
I'm very new to this field so I still have a lot of confusions in many parts.
i) Thank you for a source. Yes, I totally agree with you on this. The outcomes from CH22/CMAP and CH36 should be similar but is not necessary to be identical. However, I still think the the RMSD value differences of 0.1~0.2 is kind of big relative to the RMSD scale I got from my simulation. This is something I'm not sure if I can say they are still similar.
ii) I honestly wasn't paying too much attention on those two different versions of CH36, so thank you for bringing that up here. I used the July 2020 version of CH36, and I actually didn't specify which one there. So I must be using the default ff and even after I went over the website below, I'm still not quite sure which one is default in July 2020 (I'm guessing since they recommend to use ch36m, the default is ch36m here?). I have to check it carefully.
http://mackerell.umaryland.edu/charmm_ff.shtml
iii) Actually, I also compared the RMSF (protein, residues) and H-bonds number as well. For RMSF, while the residues that fluctuate totally overlaps, the degree of fluctuation is much high in CH36 based simulation compared to the CH22/CMAP based one. For H-bonds, the average H-bonds count is ~6150 in CH36 based one while it is ~5800 in CH22/CMAP one.
Yes, being closer to the starting structure is not necessarily equal to the better result. You are right about that. However, what I kind of see in the output structure in CH36 based protein is that there is a clear deformation of its shape after 100ns of the production run where RMSD starts showing the stableness. Since I'm really not an expert of this field nor biology/chemistry, I'm not totally sure if this is something I can get over it since RMSD is stable or it's something I need to take a serious caution since the shape has changed a lot. Again, our protein is not in an extreme condition, but in where the temperature is around 297K , pressure is 1atm, no charge and solvated box with water.
iv) I have actually compared with another person who works close to me, and both results showed that the CH36 gives higher RMSD, bigger fluctuation in RMSF and larger counts of H-bonds compared to CH22/CMAP. There are two more people who have done CH22/CMAP based simulations for the same protein under the same setting, and all the RMSD, RMSF and H-bonds count matched with my and another person's CH22/CMAP simulation data (with allowable error bars)
But thank you for suggesting the point of stochastic nature though. Yes, it is definitely dangerous using one data and concludes something here.
Best,
Karin
- Badges
- Science topic