Re: [gmx-users] free energy: annihilation
David, Yuguang reminds me of something else... I just did a bunch of hydration free energy calculations for a large set of small molecules. These tend to be pretty accurate for stuff that is mostly nonpolar, but the larger the charges are, the bigger the deviation from experiment can be. In particular, I tested a bunch of different charge models; the average error is correlated with the dipole moment, among other things. Generally, I would say that the larger your hydration free energies are, the worse you will probably do compared to experiment, unless you have some way of getting really terrific partial charges or something. If DMSO has high partial charges, it seems quite possible that the right answer for the hydration free energy with those partial charges may end up being quite different from experiment. This seems weird, the only physical reason I can think of would be a contribution due to depolarization (as in Berendsen's SPC/E model). For water this would be roughly 4 kJ/mol. However for DMSO in my case it would be less, since it has a high dipole in the gas phase as well. Obviously if the model is poor results will be poor. No, not systematic. Just think of it this way: If you have only small partial charges, the only thing that can possibly be wrong that will make much difference is the LJ parameters or the water model. If the partial charges are larger, they can also make a big difference in the hydration free energy. Thus, for mostly apolar molecules, you can't do *too* badly, since water molecules and LJ parameters are generally decent. But for polar molecules, if your partial charges are really bad, you can do very badly. In other words, all I'm saying is basically that the larger the hydration free energy is, the more wrong it can, which is fairly obvious. :) I actually found that, for my test set, depending on the partial charges I used, I could get hydration free energies for polar molecules that varied by more than a factor of 2. And these were all sensible charge sets in some sense (QM-based, fit using RESP). The moral of the story is that basically you should be careful what charges you use, and be aware that it is possible for the charge set to mess things up a whole lot. David Could there be some other systematic problem? -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED][EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] free energy: annihilation
David Mobley wrote: David, Yuguang reminds me of something else... I just did a bunch of hydration free energy calculations for a large set of small molecules. These tend to be pretty accurate for stuff that is mostly nonpolar, but the larger the charges are, the bigger the deviation from experiment can be. In particular, I tested a bunch of different charge models; the average error is correlated with the dipole moment, among other things. Generally, I would say that the larger your hydration free energies are, the worse you will probably do compared to experiment, unless you have some way of getting really terrific partial charges or something. If DMSO has high partial charges, it seems quite possible that the right answer for the hydration free energy with those partial charges may end up being quite different from experiment. This seems weird, the only physical reason I can think of would be a contribution due to depolarization (as in Berendsen's SPC/E model). For water this would be roughly 4 kJ/mol. However for DMSO in my case it would be less, since it has a high dipole in the gas phase as well. Obviously if the model is poor results will be poor. No, not systematic. Just think of it this way: If you have only small partial charges, the only thing that can possibly be wrong that will make much difference is the LJ parameters or the water model. If the partial charges are larger, they can also make a big difference in the hydration free energy. Thus, for mostly apolar molecules, you can't do *too* badly, since water molecules and LJ parameters are generally decent. But for polar molecules, if your partial charges are really bad, you can do very badly. In other words, all I'm saying is basically that the larger the hydration free energy is, the more wrong it can, which is fairly obvious. :) It still seems strange, if one can get a lot of other properties correct, like DHvap, dielectric constant, density, diffusion constant etc. Unfortunately my simulations are converging very slowly, but I hope to have more conclusive results soon. I actually found that, for my test set, depending on the partial charges I used, I could get hydration free energies for polar molecules that varied by more than a factor of 2. And these were all sensible charge sets in some sense (QM-based, fit using RESP). Reasonable Estimate of Simple Point charges... It's not that I know of any better method to do it, but the QM levels of theory typically used for computing charges are *very* low, and far from convergence to the basis-set limit. In addition RESP is not a good method, because the problem is underdetermined... You know, Peter Kollman once told me that they did quantum calculations followed by RESP and then multiplied the charges by 0.9. The moral of the story is that basically you should be careful what charges you use, and be aware that it is possible for the charge set to mess things up a whole lot. cynical Or: for interesting molecules you're out of luck, unless you're lucky /cynical -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED] [EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] free energy: annihilation
David, It still seems strange, if one can get a lot of other properties correct, like DHvap, dielectric constant, density, diffusion constant etc. Unfortunately my simulations are converging very slowly, but I hope to have more conclusive results soon. Ah, you checked all those things? I suspect in my case I would get incorrect DHvap, etc depending on my charge set. Yes, I would expect these to be pretty slow to converge: You're dealing with a pretty big molecule. I run 5 ns for relatively small molecules that don't even have any rotatable bonds. Reasonable Estimate of Simple Point charges... It's not that I know of any better method to do it, but the QM levels of theory typically used for computing charges are *very* low, and far from convergence to the basis-set limit. In addition RESP is not a good method, because the problem is underdetermined... You know, Peter Kollman once told me that they did quantum calculations followed by RESP and then multiplied the charges by 0.9. So we have a preprint on this. We've done about 8 different levels of QM going up to MP2/cc-PVTZ and B3LYP/cc-PVTZ, with and without reaction field treatment of solvent. I agree that RESP is not ideal, but I haven't seen anything that works better. If you have, I'd love to hear about it. We're not, however, doing anything strange like multiplying by 0.9. cynical Or: for interesting molecules you're out of luck, unless you're lucky /cynical Exactly. :) Or, in other words, we should worry about this far more than we do. David ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] free energy: annihilation
David Mobley wrote: Yuguang reminds me of something else... I just did a bunch of hydration free energy calculations for a large set of small molecules. These tend to be pretty accurate for stuff that is mostly nonpolar, but the larger the charges are, the bigger the deviation from experiment can be. In particular, I tested a bunch of different charge models; the average error is correlated with the dipole moment, among other things. Generally, I would say that the larger your hydration free energies are, the worse you will probably do compared to experiment, unless you have some way of getting really terrific partial charges or something. If DMSO has high partial charges, it seems quite possible that the right answer for the hydration free energy with those partial charges may end up being quite different from experiment. This seems weird, the only physical reason I can think of would be a contribution due to depolarization (as in Berendsen's SPC/E model). For water this would be roughly 4 kJ/mol. However for DMSO in my case it would be less, since it has a high dipole in the gas phase as well. Obviously if the model is poor results will be poor. Could there be some other systematic problem? -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED] [EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] free energy: annihilation
David, I'm trying to annihilate a complete DMSO box at constant volume. The dG/dlam values converge nicely but when I integrate the curve I get a Helmholtz energy that is almost exactly a factor of two different from the experimental value. Maybe this is too basic, but are you doing the transformation in vacuum also and subtracting? I assume you are trying to compute something like a hydration free energy, which means you have to complete a thermodynamic cycle that takes you from the full molecule in water to the full molecule in vacuum. I don't know any way to do this in GROMACS currently without also doing a separate vacuum calculation, because of the intramolecular interactions. Am I forgetting something here? Should I treat the intramolecular degrees of freedom in a special manner? You can, but you don't have to. One thing that can sometimes help convergence is to only turn off the intermolecular LJ interactions, which you can do by modifying the pairs list and explicitly stating appropriate LJ parameters, or (presumably) by using the new pairs types that Berk implemented, although I still haven't tested these myself. Currently I am only turning off the LJ and Coulomb terms. Should I turn off the angles and dihedrals as well? No, you don't want to do that. The other thing that can make some difference is your protocol. In particular: (1) Are you turning off the LJ and Coulomb terms simultaneously? I've always found that it is a lot harder to converge this calculation and get reliable answers than if you do the Coulomb interactions first and then do the LJ interactions. (2) What soft core parameters are you using for the LJ interactions? Some work much better than others. (My free energy tutorial has some information about doing the LJ transformation. If you want to do something really basic, you could try reproducing my value for methane: http://www.dillgroup.ucsf.edu/group/wiki/index.php/Free_Energy:_Tutorial. I also list the soft core parameters that are optimal there.) (3) Make sure your PME parameters are good enough; sometimes free energy calculations can be pretty sensitive to this sort of thing (4) How curved is your dV/dlambda plot? TI can end up making a nontrivial amount of integration error, especially if you have nearly equal positive and negative areas under the curve (in which case the integration error could be on the same order of magnitude as the total integral). David David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED][EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
RE: [gmx-users] free energy: annihilation
Hi David, Recently I repeated some free energy solvation calculations according to the tutorial by David Mobly. I agree with him that when you do DMSO which has some high partial charges, You should do calculations with Coulomb and VdW separately. Otherwise unphysical results appear. Best regards Yuguang Dr. Yuguang Mu Assistant Professor School of Biological Sciences Nanyang Technological University 60 Nanyang Drive Singapore 637551 Tel: +65-63162885 Fax: +65-67913856 http://genome.sbs.ntu.edu.sg/Staff/YGMu/index.php -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of David Mobley Sent: Saturday, August 26, 2006 9:05 PM To: Discussion list for GROMACS users Subject: Re: [gmx-users] free energy: annihilation David, I'm trying to annihilate a complete DMSO box at constant volume. The dG/dlam values converge nicely but when I integrate the curve I get a Helmholtz energy that is almost exactly a factor of two different from the experimental value. Maybe this is too basic, but are you doing the transformation in vacuum also and subtracting? I assume you are trying to compute something like a hydration free energy, which means you have to complete a thermodynamic cycle that takes you from the full molecule in water to the full molecule in vacuum. I don't know any way to do this in GROMACS currently without also doing a separate vacuum calculation, because of the intramolecular interactions. Am I forgetting something here? Should I treat the intramolecular degrees of freedom in a special manner? You can, but you don't have to. One thing that can sometimes help convergence is to only turn off the intermolecular LJ interactions, which you can do by modifying the pairs list and explicitly stating appropriate LJ parameters, or (presumably) by using the new pairs types that Berk implemented, although I still haven't tested these myself. Currently I am only turning off the LJ and Coulomb terms. Should I turn off the angles and dihedrals as well? No, you don't want to do that. The other thing that can make some difference is your protocol. In particular: (1) Are you turning off the LJ and Coulomb terms simultaneously? I've always found that it is a lot harder to converge this calculation and get reliable answers than if you do the Coulomb interactions first and then do the LJ interactions. (2) What soft core parameters are you using for the LJ interactions? Some work much better than others. (My free energy tutorial has some information about doing the LJ transformation. If you want to do something really basic, you could try reproducing my value for methane: http://www.dillgroup.ucsf.edu/group/wiki/index.php/Free_Energy:_Tutorial . I also list the soft core parameters that are optimal there.) (3) Make sure your PME parameters are good enough; sometimes free energy calculations can be pretty sensitive to this sort of thing (4) How curved is your dV/dlambda plot? TI can end up making a nontrivial amount of integration error, especially if you have nearly equal positive and negative areas under the curve (in which case the integration error could be on the same order of magnitude as the total integral). David David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED][EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php