I have now made some significant changes in the arrangement of the simulation details part of the manuscript: the simulation details texts are now in the Supplementary section and references to the simulation files and details are added to the tables describing the simulated systems. There is also updated ToDo list. For these reasons I have now made a pdf with new file name:
https://www.dropbox.com/s/us0d805lpg53kqe/HGmodel_draft2.pdf?dl=0
You can still access the previous draft version using the links in previous post.
There are also some changes in the content (marked with bold) compared to the previous version.
The goal is to submit the manuscript to the arXiv server latest at 14.1.2015 due to the grant application deadline.
Important note:
I think that it would be very useful if we would share as much simulation files as possible. Since Zenodo allows 2 GB files, also the trajectory sharing is possible. I have created the nmrlipids community in the Zenodo where some simulation files with trajectories related to this project are already shared. The more we get trajectories and other files there the better. To exemplify the potential utility of this collection, I wrote a script which automatically dowloads simulation data from Zenodo and calculates the form factor (FFcomp.sh, see also discussion with Peter Heftberger). Note that Zenodo gives a doi indentification for your data, thus it can be cited. In the current version of the manuscript these citations are used in the simulation system tables to refer the location of files. If someone does not want or cannot share the files, we can just leave these table items empty.
Here's some information (with the new numbering):
ReplyDelete14: The files for the Kukol simulations are available at: http://dx.doi.org/10.5281/zenodo.13393
18: The files for the Ulmschneider&Ulmschneider simulations are available at:
http://dx.doi.org/10.5281/zenodo.13392
25: Is this the original 'full hydration' data? If so, then it's the same system as in Table I and the values can be taken from there.
26–30: The amounts of waters per lipid are: 50, 25, 20, 15, 10 and 5. Therefore the total amounts of waters are 14400, 7200, 5760, 4320, 2880 and 1440. The production simulations lasted 100 ns. I now realized that I haven't instructed Joona Tynkkynen (who simulated and analyzed the systems) to skip some data at the beginning of the trajectories when the box size is still changing due to removal of waters. I will recalculate the values asap allowing the simulation boxes to properly equilibrate before data collection. I will also recalculate the P–N vector data.
36–38: The systems have 50 waters per lipid (either PC or cholesterol) for a total of 6400 water molecules. The simulation temperature was 310 K.
So I repeated both the P–N vector and order parameter calculations for the MacRog systems with varying water/lipid radtio discarding the first 50 ns of the trajectories. This choice is justified based on the time evolution of the area per lipid.
DeleteThe average P–N vector angles with respect to the z axis changed only very little:
w/l average angle
5 78.4
10 74.2
15 71.0
20 69.7
25 69.3
50 68.7
The order parameters are as follows:
5 waters/lipid:
0.0851 0.1279
0.1038 0.0688
-0.1898 -0.2612
-0.1881
-0.1624 0.0030
10 waters/lipid
0.0741 0.0766
0.0562 0.0550
-0.1914 -0.2639
-0.2094
-0.1771 0.0036
15 waters/lipid
0.0688 0.0503
0.0228 0.0506
-0.1450 -0.2407
-0.1991
-0.1722 0.0214
20 waters/lipid
0.0534 0.0468
0.0188 0.0335
-0.1442 -0.2426
-0.2052
-0.1746 0.0277
25 waters/lipid
0.0571 0.0306
-0.0011 0.0424
-0.1169 -0.2430
-0.2077
-0.1566 0.0102
50 waters/lipid
0.0586 0.0360
0.0063 0.0404
-0.1379 -0.2425
-0.2018
-0.1605 0.0164
Sorry for the inconvenience and the extra work.
The trajectories (xtc) and run input files (tpr) for the full hydration MacRog simulation are at:
Deletehttp://dx.doi.org/10.5281/zenodo.13497
Note that the data is saved every 100ps due to the file size limitations. The analysis was performed on data saved every 10ps so the results are not perfectly identical.
The same types of files are available for the dehydration MacRog simulations at:
http://dx.doi.org/10.5281/zenodo.13498
The force field parameters will be made available in a near-future publication. However, the uploaded files are adequate for repeating the analyses with Gromacs versions 4.6. and above.
I also noticed that the 50 waters/lipid system was actually simulated for 90 ns only. Therefore the analysis performed on 40 ns instead of 50 ns (50 ns is always discarded). This can be indicated in the Table and mentioned in the simulation description in the SI.
The order parameter values for other hydrogen in the alpha carbon with higher hydration levels is quite low in the data here. It is here 0.0063 and -0.0011 leading to significant forking. In previously reported data from MacRog model with full hydration this value has been larger (close to 0.05). Could there be some mistake here?
DeleteI am referring above to the hydration levels of 50 and 25 waters/lipid having values 0.0063 and -0.0011, respectively.
DeleteRight. Our code outputs alpha before beta and happened to forget this. Therefore the alpha and beta values have changed places in my message. Sorry for this.
DeleteI have now updated these results in the figures in the current version of the manuscript. As expected, the differences to the previous results are very minor.
DeleteIn order to give credit to the right people, I should mention that the trajectory which we refer to as the "full hydration" MacRog (http://dx.doi.org/10.5281/zenodo.13497) was obtained from Tomasz Rog. This structure and his files were also employed in constructing the systems for the dehydration and ion concentration simulations so he probably deserves to be included in the author list.
Delete1/2. Here are my comments on the manuscript. The point and section numbers below refer to the second version of the manuscript (posted on December 23).
ReplyDelete-> Point 1
I agree it would be nice to change the name of what we call "Berger lipids" to what they really are. Since they can be downloaded on Peter Tieleman's website, and that except DPPC all the other lipids refer to Peter's work, don't you think it would be better to call them Berger-Tieleman (or Berger-T) rather than Berger-M (refering to Marrink?). If we make the change in the manuscript we have then to add a statement explaining that clearly (and also we shall change our habits and cite Berger-T in our future papers!).
-> Point 2
In the added paragraph, should the link to the GitHub organization point to https://github.com/NMRlipids (rather than the actual link to the blog)?
-> Point 39
Good idea to make a separate figure for the experimental values since they somehow get lost in Fig 2 (thus only light blue areas could be left there and the dotted line taken out).
-> Point 41
I think it would be nice to have the signs on Fig 2. I don't have time to do it now and I realize it's probably a lot of work.
-> Point 44
I think this table would be useful for people trying to reproduce the data, or as a reference for future improvement.
-> Point 49
I think Luca's suggestion to increase the size in the y-axis is a good idea. Maybe we should also remove the lines between the points, especially when there are only two points. About the comment http://nmrlipids.blogspot.fr/2014/11/new-version-of-manuscript.html?showComment=1416863341375#c3991863460709460356 from Markus, I'm not sure I understand the goal of what you propose. Do you want to show the relative changes only? If so, I think we still should have one figure with the absolute values, but it's a good idea to also see if the relative trend is reproduced.
-> Point 51
The essential conclusion is there, I think a Fig in SI is fine.
-> Point 62
Were the CHARMM36-UA performed with GROMACS?
-> Extra comments on Section B. Analysis (p2-3)
- I would add a very quick explanation (one sentence) about the sign of quadrupolar splittings (especially the fact that it can't measured directly from the 2-H NMR experiment but can be assigned using some models, and on the other hand it's directly available from simulations) with a link to the blog. I suggest this because it is probably not obvious to people not confortable with 2-H NMR details (and so we were at the beginning of the blog!), but it's a really important matter.
- Ref [66] points to question marks, what is this ref?
Here are thoughts about some of your points:
DeletePoint 1: Berger-Tieleman or Berger-T are definitely better than Berger-M. (Also DPPC is Tieleman's work since he is second author in that paper.)
Point 2: This is now fixed in the current version.
Extra comments: I agree that we should mention the sign issue and refer to the blog post:
http://nmrlipids.blogspot.fi/2014/04/on-signs-of-order-parameters.html
I started to think that maybe we should change the name of Section II.B. Analysis to: "Analysis and terminology of order parameters from experiments and simulations" or similar, and also move the "forking" explanation to that section (now in section III.A.).
Ref [66] is: A. Vogel and S. Feller, The Journal of Membrane Biology
245(1), 23 (2012). Fixed in the current version.
Point 49.
DeletePatrick, you got my point right, that is, I wish to show the relative trends. This would be done by matching the y-values (experimental values on y1, simulations on y2) at full hydration.
This matching will of course make the simulations appear too good. The absolute values are given in Fig. 2, but unless the reader really reads the whole paper and thinks, they might be fooled the think that things are way better than they are. To try to avoid this confusion I am thinking of implementing four differently colored y2-axis, one for each force field.
2/2. Here is my contribution about Poger simulations.
ReplyDelete-> Point 13
Table I for line Poger[73]:
lipid Nl Nw T(K) t-sim t-an
DPPC 128 5841 323 100 50
I am fine to share the traj files, I just need to open an account on Zenodo.
Meanwhile, the input files are here: http://www.dsimb.inserm.fr/~fuchs/project_Samuli/Poger_DPPC/.
-> Point 61 (text to add in SI f section)
The Poger lipids are derived from GROMOS G53A6 [DOI: 10.1002/jcc.21396] and were initially coined 53A6-L (L for lipids), and are now part of GROMOS G54A7 [dx.doi.org/10.1021/ct300675z]. They work with the SPC water model. The initial hydrated bilayer structure of 128 DPPC/5841 water molecules as well as force field parameters were downloaded from David Poger's web site (http://compbio.chemistry.uq.edu.au/~david/) on April 2012. MD Simulations were run for 100 ns using a 2 fs time step and the analysis was performed on the last 50 ns. Coordinates were saved every 50 ps for analysis. All bond lengths were constrained with the LINCS algorithm. Temperature was kept at 323K with the v-rescale thermostat with a time constant of 0.1 ps (DPPC and water coupled separetly). Pressure was maintained semi-isotropically at 1 bar using the Parrinello-Rahman barostat using a 4 ps time constant and a compressibility of 4.5e-5 bar^-1. For non-bonded interactions, two conditions were tested:
i) A 0.8-1.4 nm twin-range cutoff with the neighbor list updated every 5 steps for both electrostatics and Lennard-Jones. For the former the generalized reaction field (RF) with a dielectric permitivity of 62 was used beyond the 1.4 nm cutoff. This is the original setup that Poger et al. used [DOI: 10.1002/jcc.21396].
ii) PME electrostatics with a real space cutoff of 1.0 nm, a Fourier spacing of 0.12 nm and an interpolation order of 4, LJ computed with a 1.0-1.4 nm twin-range cutoff, neighbor list updated every 5 steps. Note that Poger and Mark tested the effect of PME vs RF in the ref [dx.doi.org/10.1021/ct300675z], but used a 1.0 nm cutoff for LJ interactions. Since there is no dispersion correction in GROMOS (and since dispersion correction should not be used anyway for systems with interfaces), we decided to stick to the same LJ cutoff as that with the RF scheme i), that is 1.4 nm.
Since Poger lipids come from GROMOS force field, it is important to note that GROMOS uses the RF scheme for computing electrostatics (this is the method used for the force field parameterization). Using setup i) based on RF, we were able to reproduce the results (i.e. area per lipid~0.63 nm^2) from the original work only with GROMACS version <= 4.0.* (the original authors used GROMACS version 3.3.3 [DOI: 10.1002/jcc.21396]). On going to versions >= 4.5.*, the area per lipid dropped below 0.58 nm^2. The GROMACS developers were contacted and a redmine issue opened (http://redmine.gromacs.org/issues/1400). The difference comes from the new Trotter decomposition introduced in version 4.5. A fix has been introduced in version 4.6.6 that allows a recovery of ~0.615 nm^2. The results in terms of area per lipid using the different GROMACS versions GROMACS are here: http://www.dsimb.inserm.fr/~fuchs/project_Samuli/Poger_DPPC/tests_gmx_versions/)[Samuli, you may want to add this Figure+Legend+Comment to a stable link (Figshare?), let me know]. Of note, some tests (Miguel Machuqueiro, personnal communication) with the same RF scheme i) on simulations with proteins (thus with explicit hydrogens) showed that this setup is still unstable and eventually leads to crashes with versions 4.6.6 and 4.6.7 (and so does version 4.5.*).
Thus we decided to use only the PME setup ii) for computing the order parameter since it gives stable results whatever the GROMACS version. We obtained an area per lipid of ~0.615 nm^2, below 0.648 nm^2 found by the original authors with their PME setup (see dx.doi.org/10.1021/ct300675z). We explained that by the fact we used a 1.4 nm for the LJ cutoff and they used 1.0 nm.
I have now included this in the current version (found here:
Deletehttps://www.dropbox.com/s/us0d805lpg53kqe/HGmodel_draft2.pdf?dl=0)
I made some changes, please check that you agree.
I think that it would be nice to get your area per molecule comparison document under permanent link. For me it does not matter where it is, as long as the link is permanent and citable. However, I think that first it would be good to have the files under permanent link since you are referring those in the pdf document.
Should we also shortly mention here that the version of Poger we used (and that was used in their original publication) has those too dihedrals that went missing somewhere between Apr 2012 and Oct 2013?
DeleteI am referring to Patrick's comment on 25 Oct 2013: http://nmrlipids.blogspot.com/2013/10/welcome-if-you-are-new-here-reading.html?showComment=1382718128134#c3341186425139768422
Also we should mention that we did two sets of Poger simulations, with independent initial velocities.
DeleteIn Table 1 we could, e.g., have the simulation & analysis times as 2x100 & 2x50. And in the SI section f we could write: "MD Simulations *(two repetitions with independent initial velocities)* were run for 100 ns using a 2 fs time step and the analysis was performed on the last 50 ns." The part between the ** is new.
Thanks for the feedback.
Delete-> Samuli's comment:
I agree with your changes to the text. I've made an account on Zenodo and will upload the trajectories there next week, so that I'll be able to add permanent links in the PDF and will finally upload that PDF on Zenodo as well. It's better to upload everything on Zenodo since I'm not sure my web-page URL will be active in, say, 10 years. Will make you know once done.
Finally, I see that we do not cite the ref of SPC that is first encountered in Poger et al. section, but mentionned again in Kukol section. Shall we cite the ref (Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans, J. In Intermolecular Forces; Pullman, B., Ed.; Reidel Publishing Company: Dordrecht, The Netherlands, 1981, pp 331-342.)? I also noticed we don't have cited the generalized reaction field ref. I suggest to add the ref (http://dx.doi.org/10.1063/1.469273) at the end of the sentence "For the former the generalized reaction field (RF) with a dielectric permitivity of 62 was used beyond the 1.4 nm cutoff."
-> Markus' comment #1:
You're right it's important to mention this story of dihedral.
After the sentence "The initial hydrated bilayer structure of 128 DPPC/5841 water molecules as well as force field parameters were downloaded from David Poger’s web site (http://compbio.chemistry.uq.edu.au/~david/) on April 2012." I suggest to add the following:
---
We noticed that the same files downloaded in October 2013 appear to lack two dihedral angles in the choline headgroup (only one dihedral of type gd_29 allowing the rotation of the 3 choline methyls) compared to the April 2012 version (3 dihedrals of type gd_29 for the 3 choline methyls). This should not affect the bilayer structure and only change the kinetics of the choline methyls rotation, however the October 2013 version has not been tested.
---
-> Markus' comment #2:
Good suggestion to mention about the two clones of simulations for each condition (RF and PME). I agree with your proposition to change Table 1 and SI. I'll upload the 4 trajectories on Zenodo and will make you know once done.
Thanks KnQMe_4NuspqF2dqnJfBC86JFgANZmaEqZg-,
DeleteI have now included these changes into the manuscript.
I uploaded on Zenodo all the files dealing with simulations using Poger's lipids (GROMOS 53A6_L). Here are the doi:
DeleteWith reaction field electrostatics
1) Traj RF: http://dx.doi.org/10.5281/zenodo.14592
2) Traj RF2: http://dx.doi.org/10.5281/zenodo.14591
--> This is the traj I tested with different GROMACS versions. The pdf showing the results is located on the same Zenodo record (direct link to it: https://zenodo.org/record/14591/files/Comments_area_different_gmx_versions.pdf).
With PME electrostatics
3) Traj PME: http://dx.doi.org/10.5281/zenodo.14594
4) Traj PME2: http://dx.doi.org/10.5281/zenodo.14595
--> this is the two traj that were used in Fig 2
Apart from that, Fig 2 is really nice with signs included. One tiny typo I noticed in the current version of the manuscript (about my affiliation p1): it's currently written "Institut Jacques Monod, CNRS, Universit Paris Diderot, Sorbonne Paris Cit, Paris, France" because the french accents do not show up. If it's not possible to write "Université" and "Cité" with the french acute accent, we may write "Institut Jacques Monod, CNRS, Universite' Paris Diderot, Sorbonne Paris Cite', Paris, France".
About Fig. 2 and different results: As we have discussed, the Fig. 2 contains huge amount of data and might be difficult to read. For Berger and CHARMM36 we have now data for different lipids in different temperatures (see Table 1). If we would report only the results for POPC at 298K for these force fields, it would reduce the amount of data in Fig. 2 and might clarify it. On the other hand, this data would be enough for the conclusions made in the manuscript. What do you think?
ReplyDeleteThanks for pointing this out, this is a possible option.
DeleteOther things I am now trying is increasing the height-to-width ratio (brings points apart), including signs (more scatter between FFs), leaving just the blue regions for the FF-comparison (as suggested by Patrick 31 Dec 2014), and sorting the points for each carbon in a descending order ('calmer' appearance to human eye).
Let us see how congested the final plot becomes, and if it still looks too fuzzy, let's then consider dropping those, agreeably somewhat superfluous, data.
Hi,
ReplyDeleteAbout points 8, 10, 33, 34, 35 : I plan to share the xtc files and tpr in zenodo (hopefully before the 14th).
33 : Number of water molecules : 4960
34 : Number of water molecules : 4496
55: The starting structures for the pure POPC and DOPC simulations was taken from the Slipids website. (http://people.su.se/~jjm/Stockholm_Lipids/Downloads.html). Thus, We need to cite their papers.
56: Yes, they are, here the link : http://www.gromacs.org/@api/deki/files/184/=charmm36.ff_4.5.4_ref.tgz. It is also written as a README in the github repository in POPCcharmm/POPCchol
57: I used PyTopol (https://github.com/resal81/PyTopol). More precise informations can be found in comments inside the topology files (wich are avaiables in the github repo).
Hi,
DeleteMy simulations files are now on Zenodo :
POPC Charmm36: http://dx.doi.org/10.5281/zenodo.14066
POPC 20% Chol Charmm36: http://dx.doi.org/10.5281/zenodo.14067
POPC 50% Chol Charmm36: http://dx.doi.org/10.5281/zenodo.14068
Hi,
ReplyDeleteHere are the details of CHARMM36UA simulations.
Input files are available http://dx.doi.org/10.5281/zenodo.13821 .
Values for Table 1:
CHARMM36-UA [87, 88] DLPC 128 3840 323 30 20 [http://dx.doi.org/10.5281/zenodo.13821] SI
Simulation details:
A hydrated bilayer consisting of 128 DLPC lipids and 3840 water molecules is modeled by the force field of Lee and co-workers [88], which is a combination of the all-atom CHARMM36 force-field and the united-atom Berger model. The nonbonding interactions are calculated using an atom-based switching function with inner and outer cutoffs of 0.8 and 1.2 nm [88]. Long range electrostatic interactions are implemented using the particle-particle particle-mesh solver with a relative accuracy of 0.0001. The system is first equilibrated for 30 ns in the NP$\gamma$T ensemble (Nose-Hoover style thermostat and barostat with anisotropic pressure coupling) at 323 K and 1 bar with timestep of 1 fs, the next 20 ns of dynamics are taken for calculation of configurational averages. Simulations were carried out by using the LAMMPS package ( http://doi.org/10.1006/jcph.1995.1039 ), the input files are available ( http://dx.doi.org/10.5281/zenodo.13821 ).
Note that, if I understand correctly the simulation details given above by Alexandru, in Table 1 we should write t_sim(ns) = 50 and t_anal(ns) = 20.
DeleteNo, t_sim(ns) = 30 and t_anal(ns) = 20.
DeleteSo in this case it should read "The system was simulated for 30 ns in the NP$\gamma$T ensemble (Nose-Hoover style thermostat and barostat with anisotropic pressure coupling) at 323 K and 1 bar with timestep of 1 fs, the last 20 ns of the simulation was used for calculation of configurational averages."
DeleteRight?
Sorry, Markus is right. t=30 ns for equilibration and 20 ns for calculations. t_sim(ns) = 50 and t_anal(ns) = 20.
DeleteI've got one comment on the text. At the moment the manuscript says:
ReplyDelete"For example, it has been suggested that the surrounding lipids shield cholesterol from interactions with water by tilting their head groups (“umbrella model”)"
However, the original umbrella model paper does not mention that some tilting of head groups is required for the umbrella effect. I don't know if this idea has surfaced due to the simulation results but somehow the text should be adapted a bit to emphasise the origins of the tilting idea.
This is direct quotation from the Umbrella model paper [Biophys. J. 76, 2142–2157 (1999)]:
Delete"When cholesterols are incorporated into a phospholipid bilayer, phospholipid headgroups provide “cover” to shield the nonpolar part of cholesterol from exposure to water. This is illustrated schematically in Fig. 10 a. Phospholipid headgroups act like umbrellas. The space under the headgroups is shared by acyl chains and cholesterols. As the cholesterol content in a bilayer increases, polar phospholipid headgroups reorient in order to provide more coverage per headgroup for the increasing fraction of cholesterol molecules, as drawn in Fig. 10 b. The headgroup umbrellas are “stretched” to provide more coverage area."
They are talking about reorienting and stretching. In principle one can imagine headgroup stretching and/or reorientation leading to the umbrella effect sketched in Fig. 10 in the original paper without tilting. Even more imagination is needed to figure out reorientation or stretching which would leave the alpha and beta order parameters practically unchanged, as observed in experiments.
Maybe we could avoid the word tilting and write:
"For example, it has been suggested that the surrounding lipids shield cholesterol from interactions with water by reorienting their head groups (“umbrella model”) [114]"
To make it evident to the reader that we are not using a straw man argument when discussing the umbrella model, a direct quote from the original paper would maybe be in order here? That is, we would write something along the lines:
DeleteFor example, the Umbrella model suggests that "phospholipid head- groups provide 'cover' to shield the nonpolar part of cholesterol from exposure to water" such that "as the cholesterol content in a bilayer increases, polar phospholipid headgroups reorient in order to provide more coverage per headgroup for the increasing fraction of cholesterol molecules" [cite Huang 1999]. In the Superlattice model cholesterol acts as a spacer for the headgroups thus increasing their entropy and dynamics…
Should we decide to do this, we probably should use a direct quote also to describe the Superlattice model.
This may be an overkill. I have changed the "tilting" to "reorienting" for now, however we can still think about this.
DeleteHere are my comments for points 59 and 60. Should I put these files on Zendo. Should I put the original AMBER files on Zendo? The Gromacs compatible files are already included with the simulations that Samuli performed using the two FF.
ReplyDelete59) The force field parameters were generated using files obtained from the Lipidbook website (http://lipidbook.bioch.ox.ac.uk/package/show/id/150.html). The conversion to Gromacs compatible formats was performed using the acpype tool (10.1186/1756-0500-5-367). The accuracy of the conversion was check by calculating the total energy of a single POPC lipid molecule using the sander program which is part of the AmberTools14 (10.1002/wcms.1121) package and Gromacs 4.6.5. A difference of 0.002 kcal/mol was obtained between the two programs.
60) The Amber compatible force field parameters were generated using the tleap program which is integrated in the AmberTools14 package (10.1002/wcms.1121). A workflow similar to the one used previously for the conversion and validation of the GAFFLipid parameters was followed here. As before, a negligible energy difference of 0.003 kcal/mol was obtained between the two programs.
About putting to Zenodo: Do you mean the files used for Amber to Gromacs conversion? This may be reasonable, if they can be used to reproduce the conversion used in the publication. Then we could cite them. As far as understand, the original Amber files were taken from the Lipidbook so they are already available, which should be enough?
DeleteLittle bit annoying thing in Zenodo is that one cannot modify or add the files in the datasets after submission. For this reason, it might be better to keep all the stuff which is somehow in progress in GitHub, and add to Zenodo only when citation is needed.
I have now added you as an author for the GAFFlipid and Lipid14 data sets in Zenodo, since you have done the force field files.
As the reproducibility-issue [1] with GAFF (too small areas-per-lipid compared to those reported by Dickson et al.) was never solved, we probably should discuss it in the Methods.
Delete[1] http://nmrlipids.blogspot.com/2013/10/welcome-if-you-are-new-here-reading.html?showComment=1381932164332#c7642197130088592128
Hi,
ReplyDeleteI've got a somewhat crazy optional idea for Fig. 2.
This won't address the issues of the print version although I think it's really, really clear considering how much data is in the figure.
So, I thought what if we also make a dynamic online-version of Fig. 2 where one could pick the force fields one wants for easier comparison? I fiddled with some data and did a proof-of-concept (contains old Github data for Gaff/lipid14 only): https://infogr.am/order-parameters
I don't have expertise to do this properly but ideally one could make a similar plot of all the data (using perhaps CSS/html5/javascript instead of ready-made websites) and upload it to the web or in the blog.
Now, the idea is to provide a link to it in the manuscript IN ADDITION to the print figure. The link could e.g. be put to the figure caption. Now I'm hoping someone here knows the web-standards well enough to make such a figure.
-Jukka
I very much like the idea of having an interactive infographic online to support the printed figure. A bonus would be that we could keep it up to date as more data is created. Unfortunately I (at least so far) lack the skills to do such a figure.
DeleteI played a little with the mpld3 library and the result looks promising http://bit.ly/1u3zF6z. The Python code used to generate the standalone html file is here https://gist.github.com/mretegan/36d8fc6a35daf1528229#file-interactive_plot-py
DeleteHi Marius,
DeleteThis looks totally awesome!
I especially like the ability to zoom in for greater detail. Also since the code is in python it is straightforward to load new order parameter data into the array as more data is created. Big thanks for this, the python version is definitely the way to go. I like it very much.
-Jukka
Hi,
DeleteTo continue with the interactive plot, there exist the tool called "plot.ly" which allow to make interactive graphs whith python,R, whatever and upload them into your account on their website. It's then really easy to share and collaborate on a graph and remove the need to host the interactive graph.
I played with your data in here the result : https://plot.ly/~HubertSantuz/38
And here the script used : https://gist.github.com/HubLot/8af8045a6e002b0a893e#file-interactive-plot-with-plotly-py
All the up-to-date data is now in GitHub:
Deletehttps://github.com/NMRLipids/nmrlipids.blogspot.fi/tree/master/DATAreportediINblog
Maybe someone wants to test making an interactive plot of it? In the repository there is also the gnuplot script I am using to create our current static pdf.
I compiled all the data and I made a plot almost indentical that yours.
DeleteHere the link : https://plot.ly/~HubertSantuz/72/lipid-force-field-comparison/
I'm still have to highlight the "sweet spots", change the marker type when errors bars are present and rearrange a little bit the legend but the proof of concept is here!
Let me know what you think.
Great, very nice! Quick thoughts that came to my mind:
Delete1) Would it be possible to allow for zooming into (or out of) the data? Automatic zooming when new data requires it would also be cool.
2) Maybe the initial figure could be just with the experimental values and sweet spots? Then the viewer could click the simulation data on. (In this case there would need to be a button for turning on all the the data on at once, so one would not need to click each one separately if one wants to see all data.)
3) A cool function (if such exist ready-made) would be if the plot would reorder the dots to maximize visibility depending on what is being shown.
But these are just quick thoughts. Great work!
All your remarks are interested and could improve the graph but unfortunately I'm not sure I can satisfy them.
DeleteSince I'm using the plot.ly library (same as gnuplot, matplotlib,mpld3,..), everything is done automatically (interactivity, plot, zoom, etc). From the documentation, your remarks are not possible right now. Just the zooming is possible : You have a series of button in the top right of the graph to play with zooming.
In the bottom left, you also have a the embed code (in html) to insert the graph in a webpage.