Friday, November 21, 2014

New version of the manuscript

First of all, I want to thank all the contributors and readers for their courage to participate to this project. I am impressed about people's willingness to contribute for such a project. I think that we have proven that this approach can be very effective in our field.

Originally one of the ideas of the project was to improve the manuscript that was published in the arXiv. I have now basically rewritten the original manuscript based on the results published in this blog. You can find the current version of the manuscript here:

 There is also the version which is constantly updated during the writing process found here:

The current manuscript covers only the results for fully hydrated bilayers, effect of dehydration and effect of cholesterol. The plan is to write a separate manuscript about the ion-lipid interactions.

There are still several open issues and missing details in the manuscript (listed in a todo list in the manuscript). Once these have been sorted out, the manuscript will be published in arXiv and also submission to a Journal will be considered.

All kinds of comments about the manuscript are welcomed from everybody (also from people who are not contributors yet). You can submit your comments by commenting this post. The *tex file of the manuscript is also available in GitHub. If you want, you can directly modify that and make a pull request to include your modifications into the manuscript.

There are some very important practical notes:

Authorship. The authorship of the manuscript will be offered for everyone who has commented the blog, according to the "on the credits post". The final decision on authorship will be made by invited individuals themselves based on their self-assessment of their scientific contribution to the project. I have sent the manuscript to the contributors few days a ago asking if their name should be included in the author list. I have currently included the names of authors who have clearly and directly answered to the question "Will your name be included in the authorlist of the current manuscript?" To become an author you have to straightforwardly answer that question through the email or by commenting this post (i.e. sending simulation details etc. is not interpreted as a yes answer to the question about the authorship).

Comments, questions and contributions. One of the key ideas of this project is that all the discussion about the scientifc content will be done publicly through this blog. This implies also to the comments, questions and contributions regarding the manuscript. Since I was asking contributor's decisions about authorship by email, I got some replies related to the content also by email. Due to the idea of having all the discussion about content publicly available, I will repeat here the comments, questions and contributions I have received by email. Ideally further discussion related to the content of the manuscript will continued by commenting this post.


1) Matti Javanainen commented (translated from Finnish):
Time constants in Gromacs are in ps, not in (ps)^-1.
I would not report real space cut-off's anymore since, at least, in the Gromacs 4.6. and newer they are optimized in the beginning of each simulation, thus the values set in mdp file do not have any practical meaning.

SAMULI: We should check from which version this optimization thing has started.

2) Antti Lamberg asked (translated from Finnish): Why do we have only absolute values in Fig. 2.? In this figure my parameters would be good, however, they have the wrong signs.

SAMULI: The real reason is that the most of the data in Fig. 2 was contributed and the figure was made before we started to look at the signs and stereospecific labeling (largely due to your contributions). We could (and it would be somewhat useful as well) to include also signs and stereospecific labeling into these results. However, it would not change the conclusions made from that figure, i.e. that only CHARMM36, MacRog and GAFFlipid are worth of more detailed studies (now taking into account only published models). For this reason my opinion is that we should not use our time and energy to include all the details into the Fig. 2. My idea is currently that we would leave the attemptions to improve the model for further publications, and I think that in this context we could discuss the issue that there may be a model which has good absolute values, but wrong signs and stereospecific labeling. Since this does not happen in practise with the published models discussed in the current manuscript we can leave this discussion out now.


A lot of information from Matti Javanainen:

Andrea Catte send the simulation details from his simulations (these contained ions, thus they are not used in the current manusript but will be relevant for the manuscript about ion-lipid interactions):

The starting DPPC lipid bilayer, which was built with the online CHARMM-GUI
(, contained 600 lipids, 30 water molecules/lipid, Na+ and Cl- ions (150 mM NaCl). The TIP3p water model was used to solvate the system. AA MD simulations of DPPC lipid bilayers were performed at ten different  temperatures (283, 298, 303, 308, 312, 313, 314, 318, 323, and 333 K) using the GROMACS  software package version 4.5.5 and the S tockholm lipids (S lipids) force field parameters for phospholipids. After energy minimization and a short equilibration run of 50 ps (time step 1fs), 100 ns production runs were performed using a time step of 2 fs with leap-frog integrator. All covalent bonds were  constrained with the LINCS  algorithm. Coordinates were written every 100 ps. PME with real space cut-off at 1.0 nm was used for Coulomb interactions. Lennard-Jones (LJ) interactions were switched to zero between 1.0 nm and 1.4 nm. The neighbour lists were updated every 10th step with a cut-off of 1.6 nm. Temperature was coupled separately for upper and bottom leaflets of the lipid bilayer, and for water to one of the temperatures reported above with the Nosè-Hoover thermostat using a time constant of 0.5 ps. Pressure was semi-isotropically coupled to the atmospheric pressure with the Parrinello-Rahman barostat using a time constant of 10 ps. The last 40 ns of each simulation was employed for the analysis of DPPC choline and glycerol backbone order parameters.

Alexander Lyubartsev sent some simulation details: 

Here are details of simulations with Hörberg et al force field:

lipid       Nl    Nw    T(K)      t-sim      t-an

DMPC        98   2700   303         75       50
POPC       128   3840   303        100       80

DMPC simulations are from ref 19 (Högberg et al)
POPC simulations are from A.L.Rabinovich, A.P.Lyubartsev J.Phys.
Conference series
510 , 012022 (2014)

Simulation details:

Time step 2fs, Ewald summation,
cut-off 14 Å, update of neighbour list each 10 steps,
Nose-Hoover thermostat,  Parrinello-Rahman barostat,
semi-anizotropic pressure coupling,
long-range isotropic pressure correction

Final note by Samuli: This suggestion in Page 7 in the manuscript:
"Interestingly, the probability of the gaughe
conformations correlates with the order parameter difference
between the
\(\beta\) and \(\alpha\) segments: the larger the gaughe fraction
the larger the order parameter difference. This suggestion together
with the results in Fig. 3 would indicate that the correct
gaughe-trans fraction for N- - -O dihedral is larger than in
GAFFlipid but smaller than in CHARMM36."

is not supported by the recent results in the blog from the new model by
Tjörnhammar and Edholm. I will rewrite this discussion.


  1. Referring to the points in red in the manuscript:

    Point 1, on open collaboration. I think
    this is worth explaining more in detail.

    2, I am not sure if it's worth
    publishing all the blog content on
    Figshare, because I suspect it would be
    a bit confusing: too much information.
    Instead, I would think it's worth
    publishing on Figshare all the data that
    are necessary to reproduce all
    simulations: all structure files,
    topologies, mdp files, for sure. Even
    trajectories, if you want (although,
    that would be really a lot of data). I
    have never heard of Zenodo, but Figshare
    starts to be well known and already has
    an agreement with PLOS journals to
    publish all data in PLOS papers; plus,
    it gives a DOI for the content. I like
    that idea.

    19/20, simulation details in a table
    sounds like a good idea. It will be
    boring/impossible to put all simulation
    details in a table, and many are
    irrelevant anyway, so I think mdp files
    should be posted in Figshare, as a
    complement to the table.

    More comments:
    figure 2 is very crowded, difficult to
    read. I understand it is difficult to
    make it better, I have no good ideas on
    how to improve it.

    Figure 3, I suppose there is some
    specific reason for having the y axis
    upside down?

    Figures 4/5, the "better" FFs show
    dihedral distributions very different
    from each other, but they all show (if I
    well understand) decent agreement with
    experiments; is this right? if it is so,
    one could conclude that matching the
    experiments is not sufficient to say
    which FF provides the best set of
    structures, right?

    Figure 6, this is also not great, as
    Markus points out. Very difficult to
    read. Maybe simply making it bigger in
    the vertical direction would improve
    things? Same for figure 8

    Figure 7, maybe this could be calculated
    also for some other FF?

    1. Point 1, on open collaboration: Do you (or somebody else) have in mind a list of points about Open Collaboration point of view which would be most relevant to bring out in the manuscript?

      Point 2, Currenly we are collecting all the files to reproduce the simulations to GitHub ( and trajectories to Zenodo ( Eventually, I think that it would be good to have everything in the same place which would be Zenodo. The advantage of Zenodo compared to the Figshare is the larger quota allowing trajectory sharing (2Gb per file in Zenodo). Zenodo will give doi as well.

      There is two reasons why I want to publish blog content somewhere (Zenodo, Figshare or similar). First, there is some very detailed discussion relevant to the content of the paper which I would not like repeat in the paper (meaning mainly these posts:
      Secondly, for historical reasons I think it would be nice to save all the discussions in the blog behind permanent link. I do not think that we can trust blogspot in this.

      I think that we have to figure out some reasonable solution for this.

      19/20, simulation details: Table would probably be nice, however, most likely I will not do it myself. If somebody is willing to do it let us know. As said above, the mdp files will be shared anyway.

      figure 2 is very crowded: It has huge amount of data. Markus has done it. I would not be able to make it even that clear. There is this suggestion in the manuscript that maybe we should make a table to clarify the conclusions from the figure (ToDo 38.) Further suggestions are welcomed.

      Figure 3, I suppose there is some specific reason for having the y axis upside down: The reason is that very often absolute values of order parameters have been discussed (also regarding the acyl chains). For this reason people have used to look figures which looks like this (or even more for corresponding figures for acyl chains). Since most values are negative, I think it is easier for people to understand the figures with inverted axis (since they look the same as absolute value figures). If there are good arguments, we can change this.

      Figures 4/5, the "better" FFs show dihedral distributions...: None of the force fields reproduces the order parameters with experimental accuracy. The definite conclusion we can do is that none of the force fields has the correct structure. If there would be a force field which would reproduce the results with experimental accuracy we could claim that it has the correct structure. If there would be two, then we could not say which one is better. Now we could add some more detailed discussion, like that MacRog is the best for g1 and g2, but CHARMM best for the aplha and beta. However, I think that we should target to a model which is correct for all, and then make more definitive conclusions.

      Figure 6: Yes this should be improved. Let me know if there are more ideas.

      Figure 7: Matti Javanainen delivered already the results for Macrog. These will be included. Then we have all the force fields here which have been subjected for more careful studies.

    2. "Figure 6: Yes this should be improved. Let me know if there are more ideas."

      My current idea is to try to make this separately for each hydrogen and scale the y-values such that they all start from the same point at full hydration. The challenge will be to still make it clear that simulational data are not quantitatively correct. I am thinking about a way to do this well.

  2. Hi,

    Here are the details regarding the Berger and Slipids results in Table 1(5. and 10. in the manuscript)

    Berger: lipid: DPPC, N_l = 72, N_W = 2864, T(K) = 323K tsim= 140 ns tanal= 100ns.
    Slipids: lipid: DPPC, N_l = 128, N_W = 3840, T(K) = 323K tsim= 30 ns tanal= 30ns.


    1. Thanks Jukka, there is one more thing: Since your results are only results for Slipids without salt, we will use those. The simulation details paragraph about Slipids was accidentally assigned to Matti Javanainen in the first draft (point 27. in ToDo list). So it would be useful if you could provide also the further simulation details.

    2. Hi,

      Since these will be the only SLipids data, I decided to run the simulation to 100ns at least. I'll try to upload the relevant files to github or zenondo in a couple of days. Below are the details for the simulation protocol I used. If needed, feel free to modify the text in any way you like to make it more fluent with the rest of the draft.


    3. Hi,

      Slipids data are now on Zenodo (DOI: 10.5281/zenodo.13287). First 50ns were used for equilibration and the order-parameter analysis was done for the following 50-150 ns.


    4. Hi Jukka,

      considering your Berger results, which force field of the Berger family did you use? As already discussed before [1], your order parameters in the glycerol region are quite different from the other Berger results we have. Could you clarify if this is because you use some other version of Berger than Berger-DPPC-98 [2], as is now stated in the manuscript?


      [2] S.-J. Marrink, O. Berger, P. Tieleman, and F. Jähnig, Biophysical Journal 74(2), 931 (1998).

    5. Hi,

      The model used was the OPLS-AA compatible version of Berger by Tieleman et al. DOI:10.1088/0953-8984/18/28/S07

      However, I had made a big mistake in the run parameters: the vdw cutoff was 2.0 nm insted of the correct 1.0 nm. This is too long for Berger and OPLS and causes artificial freezing in the bilayer resulting in wrong glycerol order parameters.

      So big thanks for noticing the discrepancies! I'll rerun the results with correct parameters and possibly also with the Berger-DPPC-98 model for comparison.

      So, one great thing about this collaborative project is that mistakes are so easily found.


    6. I should have 60ns for Berger-DPPC-98 with proper cutoffs ready by tomorrow. Files will be on Zenodo.

    7. Hi,

      New, corrected files for Berger-DPPC-98 are now at DOI:10.5281/zenodo.13934

      The simulation conditions are now as stated in the paper for Berger based models. However, the runtime was for 60ns with the order parameter analysis spanning the whole trajectory.

      After a quick glance, now the glycerol order parameters of Berger-DPPC-98 seem to more closely match those of Berger POPC so it all makes sense.


    8. Thanks Jukka, looks good!

      Probably in the final paper we should give the order parameters after skipping, say, 30 ns from the beginning... Would you mind posting these at some point too?

    9. Hi,

      Sure, here we go. Below are the order parameters from the above-mentioned DPPC trajectory analyzed from 30ns to 60ns. I didn't find an easy way of adding another file to the already uploaded Zenodo DOI so they are as plain-text here.

      @TYPE xy
      1 0.00482577468108
      1 -0.005821936945
      1 -0.00600716575815
      2 0.00603670872231
      2 0.000859946312552
      2 -0.00851070368993
      3 0.00587745107892
      3 -0.00531827053462
      3 0.00159602933226
      4 0.0298135670971
      4 0.0595413042613
      5 0.0895853417515
      5 0.111896930945
      6 -0.2519751251
      6 -0.161482750268
      7 -0.181396400932
      8 0.130892705377
      8 0.0469505878288

    10. Hi Jukka,

      You mentioned above that you had a mistake with the vdw cut-off in the DPPC simulation. Did you have the same problem also in the simulations with ions? The simulations with ions were reported here:

      I am writing the ion lipid-ion interaction paper now, and I think that the lessons from the scaling charge test and Orange force field may be useful for the people who think how the lipid-ion interactions should be improved. In scaling test, only ion properties are changed, and in Orange only lipid properties are changed, respect to the Berger.

    11. This discussion continues here:

  3. Hi,

    Here some details regarding the paper :

    Point 1, I also think we should describe the "open collaboration" : present briefly the blog, the fact that everyone is free to join, all the discussions are public, the coauthorship thing. Maybe we can take points from "Our aim" and "On credits".

    Point 2, if we publish the blog somewhere (figshare, zenodo, ...), wouldn't we 'froze' the content of the blog? What happens if the discussion continue with new comments after we published it? Differents links in the references is not sufficient? (one for the whole blog, one for specific interested comments)

    Point 19/20, we can do both, share the mdp files and make a table. On this matter, can't we just use the table I and add the simulations with cholesterol and dehydration (with a separator for the different kinds of simulations). Then, we'll have a quick overview of all simulations and not just a bunch of it. I can try to do something about it and let you know.

    1. The previous link pointed to a outdated version of the document. Here the new one :

    2. Point 2, 'Freezing' of the blog is indeed potential problem. However, I think that we should not put the links to this blog in references, since there is no guarantee how long these links will work or how long Google will keep this content online. I would assume that these links would not work, for example, ten years from now (if you are reading in 2024, I was wrong). In this respect figshare, Zenodo or some other location with unique reference is necessary. There seems to be commenting possibilty, at least, in Figshare. One option might be to use that after publication. One possibility would be also direct the after publication discussion to some specific place, for example, to a new comment in this blog.

      Point 19/20, to clarify this, the mdp files will be shared and the content will be described also in the manuscript. The question is now that if someone is willing to make a table which contains mdp file information (suggestion by Markus), or if we describe those in a traditional way as text. It is probably a good idea to put also the cholesterol and dehydration simulations into Table I (containing only the description of simulated systems). I can do this when I soon start to include the comments into the manuscript.

      About the effect of cholesterol to the dihedral angles in the attached document: Indeed the effect is small, however, I think that we could consider carefully suggesting this as an interpretation for the change in the order parameter.

    3. Hi Hubert, could you share the dihedral distributions in your pdf file in ASCII format? I will put your results into the Figure 4 and 5 (even though they seem identical to mine) and preliminary include the suggestion about the effect of cholesterol.

    4. Hi Samuli,

      Sorry for the late response.
      Here you can find the plot in ascii :
      I could but those also in the github repo if you want

    5. Thanks. I think that it would be good to get as much as possible this kind of data into the GitHub.

      One more thing, could share the figure (and/or the script used to generate it) which you have in your pdf file (different dihedrals with different cholesterol concentrations)? It would be convenient so I do not have to reproduce it from the data.

    6. By default, my figure should be in pdf. I create a script to reproduce the figure (usually, it's done within a IPython Notebook), it's in python & matplotlib.
      The script (+ the data) are in github repo.

    7. Hi,

      To re-open the discusion about the publication of the blog and it's existence in the long-run, here a great info :

      Basically, Winnower offers the possibility to get DOI for contents of blogs. It already works with wordpress hosted at and will be available shortly for Blogger blogs (i.e like this one!)

    8. The addition of blogspot blogs to the Winnower system sort of works now. However, there are some problems: Figures work only in html version (not in pdf) and comments are not included. Comments could be added manually, though. However, currently I think that easier would be to print the blog pages to pdf and add to Figshare.

    9. It seems that the most practical solution currently is the internet archive:

      Markus has put the blog there and these links are now cited in the manuscript.


Please sign in before writing your comment.