Re: [gmx-users] Bennet error in FEP-calculations for charged ligands.
Michael, Thank you so much for such a quick and detailed response. I tried to increase the number of intermediate lambda windows, but unfortunately, it was not possible to reduce the error to an acceptable level. coul-lambdas = 0.0 0.02 0.05 0.08 0.12 0.16 0.2 0.3 0.4 0.6 0.7 0.8 0.84 0.88 0.92 0.95 0.98 1.0 point 0 - 1, DG 47.68 +/- 0.13 point 1 - 2, DG 34.31 +/- 0.56 point 2 - 3, DG 19.42 +/- 0.97 point 3 - 4, DG 15.56 +/- 1.07 point 4 - 5, DG 8.45 +/- 0.03 point 5 - 6, DG 3.53 +/- 0.02 point 6 - 7, DG 4.99 +/- 0.02 point 7 - 8, DG 2.30 +/- 0.02 point 8 - 9, DG -0.02 +/- 0.03 point 9 - 10, DG -2.79 +/- 0.31 point 10 - 11, DG -5.74 +/- 0.72 point 11 - 12, DG -6.58 +/- 0.45 point 12 - 13, DG -8.02 +/- 0.03 point 13 - 14, DG -8.60 +/- 0.03 point 14 - 15, DG -15.92 +/- 0.04 point 15 - 16, DG -33.08 +/- 1.38 point 16 - 17, DG -47.34 +/- 1.22 total 0 - 17, DG 8.15 +/- 1.69 As for the three-level transformation, I agree that it should help, but it seems I don’t quite understand how to implement this behavior in GROMACS. I don't know by which options of the .mdp file it should be implemented and how to build topology files for such transformations. If you provided me with a small example of an mdp and itp file for such a transformation, I would be very grateful to you. Thank you, Shekhovtsov Artem On Wed, Jan 23, 2019 at 7:00 PM Michael Shirts wrote: > As free energies get larger, then the error gets less accurate. So if it is > reporting 18.51 +/- 2.95 for one of the intervals, that likely suggest > there is very little overlap in that area. > A total free energy difference of in -6.12 +/- 3.19 for the > transformation indicates that the result is not very certain; you're within > two standard deviations, and again, the Bennett error formula is inaccurate > for larger error. I would suggest adding 1-2 more intermediates between > those points. > > https://github.com/MobleyLab/alchemical-analysis provides some of the > tools > to check overlap. > > I would also suggest doing a multi stage transformation to increase overlap > - for atoms that are disappearing or appearing that are charged, turn off > charges off, change vdw, then turn charges back on. In some cases, > especially when the atoms involved have different sizes, sc-coul can lead > to some semi-pathological cases when vdw softcore and coul softcore result > in some very low/high potentials at intermediates. > > On Wed, Jan 23, 2019 at 8:21 AM Artem Shekhovtsov < > job.shekhovt...@gmail.com> > wrote: > > > Hi all! > > I encountered an error in my relative free energy calculations and do not > > know how to fix it. > > Molecules for which I want to carry out calculations contain a carboxyl > > group. > > To validate the protocol, I tried to run the fep-calculations of > symmetric > > molecules for which the change in free energy will be zero. > > During validation, I was faced with the fact that the convergence error > for > > charged ligands greatly exceeds that for neutral ones. > > For example, for the case of m-methyl benzoic acid [O-]C(= O)c1cc(C)ccc1: > > Mutation (-CH3 + H) for one meta position (-H + CH3) for another meta > > position relative to the carboxyl group. > > > > Ionized acid (solvent leg, 5 ns): > > point 0 - 1, DG 28.74 +/- 0.02 > > point 1 - 2, DG 53.79 +/- 0.41 > > point 2 - 3, DG 27.76 +/- 2.45 > > point 3 - 4, DG 18.51 +/- 2.95 > > point 4 - 5, DG 8.80 +/- 0.79 > > point 5 - 6, DG 4.04 +/- 0.07 > > point 6 - 7, DG -0.06 +/- 0.04 > > point 7 - 8, DG -2.62 +/- 0.33 > > point 8 - 9, DG -8.95 +/- 0.27 > > point 9 - 10, DG -23.89 +/- 1.61 > > point 10 - 11, DG -29.32 +/- 1.29 > > point 11 - 12, DG -54.15 +/- 0.08 > > point 12 - 13, DG -28.79 +/- 0.03 > > > > total 0 - 13, DG -6.12 +/- 3.19 > > > > Unionized acid (solvent leg, 5 ns): > > point 0 - 1, DG 0.08 +/- 0.01 > > point 1 - 2, DG -6.88 +/- 0.10 > > point 2 - 3, DG -14.69 +/- 0.05 > > point 3 - 4, DG -21.07 +/- 0.03 > > point 4 - 5, DG -13.21 +/- 0.02 > > point 5 - 6, DG -7.46 +/- 0.02 > > point 6 - 7, DG -0.09 +/- 0.03 > > point 7 - 8, DG 7.37 +/- 0.02 > > point 8 - 9, DG 13.24 +/- 0.03 > > point 9 - 10, DG 21.15 +/- 0.06 > > point 10 - 11, DG 14.72 +/- 0.03 > > point 11 - 12, DG 6.98 +/- 0.07 > > point 12 - 13, DG -0.05 +/- 0.01 > > > > total 0 - 13, DG 0.09 +/- 0.22 > > > > For a neutral molecule containing a charged carboxyl and amino groups > > ([O-]C(=O)c1cc(C)c([NH3+])cc1) the error is small: >
Re: [gmx-users] Bennet error in FEP-calculations for charged ligands.
As free energies get larger, then the error gets less accurate. So if it is reporting 18.51 +/- 2.95 for one of the intervals, that likely suggest there is very little overlap in that area. A total free energy difference of in -6.12 +/- 3.19 for the transformation indicates that the result is not very certain; you're within two standard deviations, and again, the Bennett error formula is inaccurate for larger error. I would suggest adding 1-2 more intermediates between those points. https://github.com/MobleyLab/alchemical-analysis provides some of the tools to check overlap. I would also suggest doing a multi stage transformation to increase overlap - for atoms that are disappearing or appearing that are charged, turn off charges off, change vdw, then turn charges back on. In some cases, especially when the atoms involved have different sizes, sc-coul can lead to some semi-pathological cases when vdw softcore and coul softcore result in some very low/high potentials at intermediates. On Wed, Jan 23, 2019 at 8:21 AM Artem Shekhovtsov wrote: > Hi all! > I encountered an error in my relative free energy calculations and do not > know how to fix it. > Molecules for which I want to carry out calculations contain a carboxyl > group. > To validate the protocol, I tried to run the fep-calculations of symmetric > molecules for which the change in free energy will be zero. > During validation, I was faced with the fact that the convergence error for > charged ligands greatly exceeds that for neutral ones. > For example, for the case of m-methyl benzoic acid [O-]C(= O)c1cc(C)ccc1: > Mutation (-CH3 + H) for one meta position (-H + CH3) for another meta > position relative to the carboxyl group. > > Ionized acid (solvent leg, 5 ns): > point 0 - 1, DG 28.74 +/- 0.02 > point 1 - 2, DG 53.79 +/- 0.41 > point 2 - 3, DG 27.76 +/- 2.45 > point 3 - 4, DG 18.51 +/- 2.95 > point 4 - 5, DG 8.80 +/- 0.79 > point 5 - 6, DG 4.04 +/- 0.07 > point 6 - 7, DG -0.06 +/- 0.04 > point 7 - 8, DG -2.62 +/- 0.33 > point 8 - 9, DG -8.95 +/- 0.27 > point 9 - 10, DG -23.89 +/- 1.61 > point 10 - 11, DG -29.32 +/- 1.29 > point 11 - 12, DG -54.15 +/- 0.08 > point 12 - 13, DG -28.79 +/- 0.03 > > total 0 - 13, DG -6.12 +/- 3.19 > > Unionized acid (solvent leg, 5 ns): > point 0 - 1, DG 0.08 +/- 0.01 > point 1 - 2, DG -6.88 +/- 0.10 > point 2 - 3, DG -14.69 +/- 0.05 > point 3 - 4, DG -21.07 +/- 0.03 > point 4 - 5, DG -13.21 +/- 0.02 > point 5 - 6, DG -7.46 +/- 0.02 > point 6 - 7, DG -0.09 +/- 0.03 > point 7 - 8, DG 7.37 +/- 0.02 > point 8 - 9, DG 13.24 +/- 0.03 > point 9 - 10, DG 21.15 +/- 0.06 > point 10 - 11, DG 14.72 +/- 0.03 > point 11 - 12, DG 6.98 +/- 0.07 > point 12 - 13, DG -0.05 +/- 0.01 > > total 0 - 13, DG 0.09 +/- 0.22 > > For a neutral molecule containing a charged carboxyl and amino groups > ([O-]C(=O)c1cc(C)c([NH3+])cc1) the error is small: > point 0 - 1, DG -4.65 +/- 0.01 > point 1 - 2, DG -19.95 +/- 0.04 > point 2 - 3, DG -28.67 +/- 0.13 > point 3 - 4, DG -57.84 +/- 0.19 > point 4 - 5, DG -44.18 +/- 0.08 > point 5 - 6, DG -26.84 +/- 0.05 > point 6 - 7, DG 0.10 +/- 0.20 > point 7 - 8, DG 26.92 +/- 0.06 > point 8 - 9, DG 44.28 +/- 0.13 > point 9 - 10, DG 57.79 +/- 0.13 > point 10 - 11, DG 28.68 +/- 0.09 > point 11 - 12, DG 19.98 +/- 0.12 > point 12 - 13, DG 4.66 +/- 0.03 > > total 0 - 13, DG 0.28 +/- 0.21 > > Adding Na and Cl ions to ([O-]C(=O)c1cc(C)c([NH3+])cc1) does not cause an > increase in error. > point 0 - 1, DG -4.61 +/- 0.01 > point 1 - 2, DG -19.96 +/- 0.07 > point 2 - 3, DG -28.68 +/- 0.04 > point 3 - 4, DG -57.67 +/- 0.04 > point 4 - 5, DG -44.24 +/- 0.01 > point 5 - 6, DG -26.91 +/- 0.04 > point 6 - 7, DG 0.03 +/- 0.03 > point 7 - 8, DG 26.89 +/- 0.03 > point 8 - 9, DG 44.20 +/- 0.08 > point 9 - 10, DG 57.75 +/- 0.09 > point 10 - 11, DG 28.65 +/- 0.06 > point 11 - 12, DG 20.02 +/- 0.05 > point 12 - 13, DG 4.67 +/- 0.01 > > total 0 - 13, DG 0.12 +/- 0.18 > > FEP-part of *.mdp: > free-energy = yes > sc-power = 1 > sc-alpha = 0.5 > sc-coul = yes > fep-lambdas = 0.0 0.01 0.05 0.1 0.2 0.3 0.4 0.6 0.7 > 0.8 0.9 0.95 0.99 1.0 > coul-lambdas = 0.0 0.01 0.05 0.1 0.2 0.3 0.4 0.6 0.7 > 0.8 0.9 0.95 0.99 1.0 > vdw-lam
[gmx-users] Bennet error in FEP-calculations for charged ligands.
Hi all! I encountered an error in my relative free energy calculations and do not know how to fix it. Molecules for which I want to carry out calculations contain a carboxyl group. To validate the protocol, I tried to run the fep-calculations of symmetric molecules for which the change in free energy will be zero. During validation, I was faced with the fact that the convergence error for charged ligands greatly exceeds that for neutral ones. For example, for the case of m-methyl benzoic acid [O-]C(= O)c1cc(C)ccc1: Mutation (-CH3 + H) for one meta position (-H + CH3) for another meta position relative to the carboxyl group. Ionized acid (solvent leg, 5 ns): point 0 - 1, DG 28.74 +/- 0.02 point 1 - 2, DG 53.79 +/- 0.41 point 2 - 3, DG 27.76 +/- 2.45 point 3 - 4, DG 18.51 +/- 2.95 point 4 - 5, DG 8.80 +/- 0.79 point 5 - 6, DG 4.04 +/- 0.07 point 6 - 7, DG -0.06 +/- 0.04 point 7 - 8, DG -2.62 +/- 0.33 point 8 - 9, DG -8.95 +/- 0.27 point 9 - 10, DG -23.89 +/- 1.61 point 10 - 11, DG -29.32 +/- 1.29 point 11 - 12, DG -54.15 +/- 0.08 point 12 - 13, DG -28.79 +/- 0.03 total 0 - 13, DG -6.12 +/- 3.19 Unionized acid (solvent leg, 5 ns): point 0 - 1, DG 0.08 +/- 0.01 point 1 - 2, DG -6.88 +/- 0.10 point 2 - 3, DG -14.69 +/- 0.05 point 3 - 4, DG -21.07 +/- 0.03 point 4 - 5, DG -13.21 +/- 0.02 point 5 - 6, DG -7.46 +/- 0.02 point 6 - 7, DG -0.09 +/- 0.03 point 7 - 8, DG 7.37 +/- 0.02 point 8 - 9, DG 13.24 +/- 0.03 point 9 - 10, DG 21.15 +/- 0.06 point 10 - 11, DG 14.72 +/- 0.03 point 11 - 12, DG 6.98 +/- 0.07 point 12 - 13, DG -0.05 +/- 0.01 total 0 - 13, DG 0.09 +/- 0.22 For a neutral molecule containing a charged carboxyl and amino groups ([O-]C(=O)c1cc(C)c([NH3+])cc1) the error is small: point 0 - 1, DG -4.65 +/- 0.01 point 1 - 2, DG -19.95 +/- 0.04 point 2 - 3, DG -28.67 +/- 0.13 point 3 - 4, DG -57.84 +/- 0.19 point 4 - 5, DG -44.18 +/- 0.08 point 5 - 6, DG -26.84 +/- 0.05 point 6 - 7, DG 0.10 +/- 0.20 point 7 - 8, DG 26.92 +/- 0.06 point 8 - 9, DG 44.28 +/- 0.13 point 9 - 10, DG 57.79 +/- 0.13 point 10 - 11, DG 28.68 +/- 0.09 point 11 - 12, DG 19.98 +/- 0.12 point 12 - 13, DG 4.66 +/- 0.03 total 0 - 13, DG 0.28 +/- 0.21 Adding Na and Cl ions to ([O-]C(=O)c1cc(C)c([NH3+])cc1) does not cause an increase in error. point 0 - 1, DG -4.61 +/- 0.01 point 1 - 2, DG -19.96 +/- 0.07 point 2 - 3, DG -28.68 +/- 0.04 point 3 - 4, DG -57.67 +/- 0.04 point 4 - 5, DG -44.24 +/- 0.01 point 5 - 6, DG -26.91 +/- 0.04 point 6 - 7, DG 0.03 +/- 0.03 point 7 - 8, DG 26.89 +/- 0.03 point 8 - 9, DG 44.20 +/- 0.08 point 9 - 10, DG 57.75 +/- 0.09 point 10 - 11, DG 28.65 +/- 0.06 point 11 - 12, DG 20.02 +/- 0.05 point 12 - 13, DG 4.67 +/- 0.01 total 0 - 13, DG 0.12 +/- 0.18 FEP-part of *.mdp: free-energy = yes sc-power = 1 sc-alpha = 0.5 sc-coul = yes fep-lambdas = 0.0 0.01 0.05 0.1 0.2 0.3 0.4 0.6 0.7 0.8 0.9 0.95 0.99 1.0 coul-lambdas = 0.0 0.01 0.05 0.1 0.2 0.3 0.4 0.6 0.7 0.8 0.9 0.95 0.99 1.0 vdw-lambdas = 0.0 0.01 0.05 0.1 0.2 0.3 0.4 0.6 0.7 0.8 0.9 0.95 0.99 1.0 What is the reason of such error and how I can fix it? mdp, itp, gro, xvg files by link - https://drive.google.com/open?id=1MiepOQb2QAZ9rclpY13owCnl8QWcTHnf Ready to provide any additional information. Thank you, Shekhovtsov Artem -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
