tag:blogger.com,1999:blog-1283194685237019772.post6307234975379107370..comments2022-08-31T14:15:18.096+03:00Comments on The NMRlipids project: NMRlipids databank: Quality evaluationmarkus.http://www.blogger.com/profile/05973383391755193687noreply@blogger.comBlogger8125tag:blogger.com,1999:blog-1283194685237019772.post-4703084799504101912022-08-31T14:15:18.096+03:002022-08-31T14:15:18.096+03:00I did more careful analysis on the size dependence...I did more careful analysis on the size dependence of form factors and order parameters using the recent dataset containing simulations in different box sizes and cholesterol concentrations: https://doi.org/10.5281/zenodo.5767450<br /><br />The results can be found from here: https://raw.githubusercontent.com/NMRLipids/DataBankManuscriptText/main/Figures/SizeDependence.pdf<br /><br />Indeed, the lobe heights seem to decrease with increasing simulation box size. This seems to be related to the reduced peak heights in the electron density profiles. Nevertherless, the locations of form factor minima and order parameters are practically independent on the simulation box size.<br /><br />In conclusion, using the full form factor in the quality evaluation (as currently done) makes the quality dependent on the simulation box size. Therefore, a quality measure focusing on locations of form factor minima would be probably better. I am currently investigating different options for this.Samuli Ollilahttps://www.blogger.com/profile/06106569992787533569noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-47750615785330902252022-05-30T14:49:06.181+03:002022-05-30T14:49:06.181+03:00I think I included the sim2exp version to the slid...I think I included the sim2exp version to the slides in Berlin. However, including simulation error in this formula is quite easy. Now xi^2 essentially gives the difference between the simulated and experiment in the "units" of experimental error. To convert the scaling to experimental+simulation error one could just replace delta_Fe with delta_Fe+delta_Fs in the xi^2 formula.<br /><br />I have not checked how the size effect is with other system sizes and if it seems to saturate at some point. As Samuli said, it would be useful to do this analysis from the databank. <br /><br />As I said above, the minima are unaffected by the size so indeed it seems (luckily) unlikely that the strong correlation of the minima with the membrane properties is obscured by the possible simulation size effect.Hannehttps://www.blogger.com/profile/05160906501508849933noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-53408973860941966762022-05-28T23:17:36.093+03:002022-05-28T23:17:36.093+03:00I do not remember such equation and do not find it...I do not remember such equation and do not find it from the notes that I have: https://raw.githubusercontent.com/NMRLipids/MATCH/master/scripts/NMRL3_analysis/qualitymeas.pdfSamuli Ollilahttps://www.blogger.com/profile/06106569992787533569noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-90201333233357486562022-05-28T13:35:48.910+03:002022-05-28T13:35:48.910+03:00Samuli wrote: "The form factor quality estima...Samuli wrote: <i>"The form factor quality estimator is currently taken directly from the SIMtoEXP and the simulation error is not there, so it is not in our equation either. If we would include it, would you have an idea how to do it exactly?"</i><br /><br />If I recall correctly, Hanne proposed an equation for this already in the NMRlipids2019 workshop in Berlin? Hanne, do you still remember how it was? Or Samuli, do you maybe still have the notes of what was proposed?markus.https://www.blogger.com/profile/05973383391755193687noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-12534255491635542262022-05-24T14:12:44.012+03:002022-05-24T14:12:44.012+03:00About the system size, 34 lipids is quite a bit sm...About the system size, 34 lipids is quite a bit smaller than typical simulations. Have you checked whether there is size dependency when comparing only systems with more than 100 lipids for example? I have been looking correlations between form factor minima, area per lipid, thickness and membrane order, and the minima seems to correlate well with membrane properties, see https://github.com/NMRLipids/DatabankExercises/blob/master/APL/correlations.pdf and https://github.com/NMRLipids/DatabankExercises/blob/master/APL/AreaPerLipidAndThicknessExamples.ipynb <br /><br />It might be useful to try to figure out from the NMRlipids databank if lobe heights correlate with some membrane properties.Samuli Ollilahttps://www.blogger.com/profile/06106569992787533569noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-70321156326972395142022-05-23T08:56:49.297+03:002022-05-23T08:56:49.297+03:00The form factor quality estimator is currently tak...The form factor quality estimator is currently taken directly from the SIMtoEXP and the simulation error is not there, so it is not in our equation either. If we would include it, would you have an idea how to do it exactly?<br /><br />Adding errors to plots is a good idea. Currently the code does not print the error, but I can take a look if we could do it. I am currently updating the code to work properly for united atom systems and systems with periodic jumps in z-direction.Samuli Ollilahttps://www.blogger.com/profile/06106569992787533569noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-39464169232296468402022-05-20T17:13:08.172+03:002022-05-20T17:13:08.172+03:00Additional reminder that the (relative) form facto...Additional reminder that the (relative) form factor lobe heights seem to be fairly sensitive to the system size. I just calculated a comparison of C36 and Slipids form factors from small (34 lipids) and big (200 lipids) simulations and scaled the first lobes to be identical for the different system sizes: <br />https://github.com/hsantila/FormFactor_data/blob/main/Slipids_C36.png<br /><br />A quality estimator were significant amount of comparison point are on the lobes is very much affected by finite size effect. Conversely, the minima and maxima location in q-axis are fairly insensitive. A quick fix might be heuristically limiting the q-range where the comparison is performed?Hannehttps://www.blogger.com/profile/05160906501508849933noreply@blogger.comtag:blogger.com,1999:blog-1283194685237019772.post-25115658489487247612022-05-20T11:43:50.555+03:002022-05-20T11:43:50.555+03:00Looks amazing! You and Anne have done very nice jo...Looks amazing! You and Anne have done very nice job with this.<br /><br />I would still consider taking account the errors in the form factors too since they are accounted for the order parameters. <br />If you definitely do not want to add them to the quality estimator, they would be nice to have at least the plots so one can visually asses the overlap with the experimental data and where the curves are most accurate. <br /><br />Error bars should be an easy addition to the code if you calculate the form factors averaging the frame-vise form factors.<br /><br />Having now found several caveats in the converge and calculation of the form factors, I have trouble trusting any curves published in the litterature. It would sense to do this with care once an for all for the databank.<br /><br />BR<br /><br />Hanne<br /> Hannehttps://www.blogger.com/profile/05160906501508849933noreply@blogger.com