- Posts: 16
Lipid self diffusion values
- pela3247
- Topic Author
- Offline
- Fresh Boarder
My apologies if this is a rather long question...
I have been running some CG simulations of different mixtures of tri-, di- and monoglycerides (with topologies obtained from the download section of the cgmartini.nl website, and then modified by changing the number of particles to get different tail lengths).
I have also included different amounts of water in my simulations, and one of the analysis I was hoping to do was to calculate self-diffusion coefficients for both water and lipids, for example to see at what water content any phase transitions occurs (phase changes often cause sharp changes in molecular mobility).
I simply use gmx msd (with the -rmcomm option) to do the calculation. For a pure box of martini water I get a reasonable agreement with the published values (0.3775e-5 cm^2*s-1, which times 4 becomes 1.51e-5 (compared to the published value of ~2.0e-5, so a bit low perhaps but still ok?)
Then, for my actual systems with water and lipids, using the same methodology, the self diffusion of the water-phase is lower (as expected, since the lipids will hinder mobility) and sort of reasonable I think (around 1e-5 at 75% (w/w) water ). For the lipids however, the diffusion values seem to be about 10 fold higher than what they should be (after applying the multiplication factor 4 to the output from gmx msd).
For example, experimentally, the diffusion coefficients of glycerol monooleate have been measured within cubic phases as 0.018 × 10−5 cm2s−1 (Ia3d cubic phase, 22% water), 0.024 × 10−5 cm2s−1 (Pn3m cubic phase, 39.5% water) and 0.02 × 10−5 cm2s−1 (Ia3d cubic phase, 81% water). I get for (roughly the same type of system) values around 0.27 *10-5 at 20% water, and 0.20 *10-5 at 75% water).
Are these values still reasonable? What would be accepted deviations for calculated values when comparing to experimental ones? Also, visually, I see the lipids forming a lamellar phase around 50% water, and so I would expect the lipids to have about the same level of self-diffusion at all water concentrations higher than this, but instead I see a monotonous decrease.
Thanks
/Per
Please Log in or Create an account to join the conversation.
- peterkroon
- Offline
- Gold Boarder
- Posts: 210
Please Log in or Create an account to join the conversation.
- pela3247
- Topic Author
- Offline
- Fresh Boarder
- Posts: 16
Thanks for your reply. No, I did not as all of these system initially have all lipids dispersed randomly in the box. So while at the end of a simulation, particularity the systems with high water content will have formed a lamellar phase, they don't initially and I wanted then to calculated the 3-d self diffusion.
Cheers
Please Log in or Create an account to join the conversation.
- peterkroon
- Offline
- Gold Boarder
- Posts: 210
Please Log in or Create an account to join the conversation.
- pela3247
- Topic Author
- Offline
- Fresh Boarder
- Posts: 16
Where you able to get someones opinion on this?
Thanks - much appreciated
/P
Please Log in or Create an account to join the conversation.
- jbarnoud
- Offline
- Admin
- Posts: 23
I cannot tell you for the cubic phase as I never had to deal with it. Regarding the lamellar phase, however, experimental diffusions are usually measured in 2D. In addition, until you form the lamellar phase, you did not reach equilibrium. I am not sure how much sense it makes to calculate the diffusion out of equilibrium.
What fraction of the MSD did you use for the fit? By default, gmx msd discards the first and last 10% of the curve; this is usually not enough to discard as it includes much more than the linear part of the curve. Using the default behaviour of gmx msd often results in underestimated diffusions.
Please Log in or Create an account to join the conversation.
- pela3247
- Topic Author
- Offline
- Fresh Boarder
- Posts: 16
Following up on this some more, since it still keeps me confused. I also now have tried to reproduce the self-diffusion constant of water using the Martini W (non-polarizable) particles
I built a system with 400 W particles, and ran for 150 ns using the mdp-file available from this site.
Then using gmx msd, I computed the diffusion constant D to be 0.38e-5 cm^2/s. This then should be compared to the "original" 2004 value of 0.5e-5.
Mutliplying by 4 (which might or might not be necessary, I take it) then gives me 1.52 e-5 cm^2/s. Is this value "close enough" to the 2.0 e-5 you get from the 2004 paper?
Thanks for your time!
/P
Please Log in or Create an account to join the conversation.
- peterkroon
- Offline
- Gold Boarder
- Posts: 210
Please Log in or Create an account to join the conversation.
- Pim
- Offline
- Expert Boarder
- Posts: 105
Please Log in or Create an account to join the conversation.
- pela3247
- Topic Author
- Offline
- Fresh Boarder
- Posts: 16
Yes, I think this value is converged, at least it's definitely a fit though the straight part of the msd-curve.
Pim: Interesting, it would be good if you would be able to report back with what you have some some point!
/P
Please Log in or Create an account to join the conversation.
- Pim
- Offline
- Expert Boarder
- Posts: 105
journals.plos.org/ploscompbiol/article?i...rnal.pcbi.1000810#s3 )
In such small boxes it is indeed much more likely that solvents freeze.
By the way, with polarizable water I find 2.14 e-5 cm2/s for a big box.
Please Log in or Create an account to join the conversation.
- pela3247
- Topic Author
- Offline
- Fresh Boarder
- Posts: 16
That's probably it. I did not think too much about freezing, but in hindsight since my simulations are run at 300, this is probably what happens (although I haven't verified 100%)
I also now did a test with a larger box with polarizable water, and got 2.16 e-5cm2/s, so I will probably switch to using this instead. Apart from some changes in the mdp-file, and the fact that the number of particles triple in the system, there are no real negative effects, right?
Please Log in or Create an account to join the conversation.
- Pim
- Offline
- Expert Boarder
- Posts: 105
If you opt for it you indeed have ~triple the particles and if you also use PME this slows you down another factor of 3-4. This is of course quite some cost but if it's not an issue I see no reason not to use it.
You can also use the polarizable force field with just reaction field electrostatics (to not lose this factor of 3-4 in speed), this is generally considered still a bit better than normal Martini, but the long-range electrostatic interactions are not accurately represented.
Please Log in or Create an account to join the conversation.