From: Assa Sittner <assaz@wi...>  20050609 22:43:03

Hi, i'm trying to calculate solvation energies of a protein in two different conformations, and i've got a few questions, and excuse me if they're stupid (today's the 3rd day i'm using apbs): 1. as i know it, and correct me if i'm wrong, i have to calculate , (for each conformation in my case), the difference in total energy energy between the "solvated protein" (using solvent dielectric) and the protein in vacuo. i did it, using different grid spacing, and received interesting results: 1. for two grids with the same size but different spacing, i have performed a solvation calculation (=totEsolvtotEvacuo), i received different absolute results for total energy(totEsolv and totEvacuo), but the solvation energies (=totEsolvtotEvacuo) stayed very close in both grid resolutions. i understand that there are artifacts, due to the discrete nature of the grid, but are they so strong? here are the figures: in both cases i used Grid lengths: 58.761 x 32.129 x 25.828 in which the protein fits very comfortably. for a 161 x 97 x 97 grid: totEsolv =9.534619136962E+04 kJ/mol totEvacuo =9.875123989039E+04 kJ/mol totEsolvtotEvacuo= 0.340504852 E+04 kJ/mol for a 161 x 129 x 129 grid : totEsolv = 6.803247450678E+04 kJ/mol totEvacuo = 7.141364011794E+04 kJ/mol totEsolvtotEvacuo=0.338116561 E+04 kJ/mol i read some posting about it, that said that the coulombic energy should be added: http://cholla.wustl.edu/pipermail/apbsusers/2003October/000104.html can anyone give me a little deeper explanation about why it's done, and how can this correction be gridresolutiondependant (since i need a different correction for each grid obviously)?is there also an effect of grid size (for a given spacing), because of the larget solvent volume? 2. what is a legitimate grid spacing (in angstroms) for calculations, in order for the to be meaningful (but not necessarily very accurate)? 3. how can i estimate the error, in the energy calculations? (except for running many calculations for different orientations on the molecule) thanks a lot in advance, T 
From: Nathan Baker <sobolevnrm@ya...>  20050613 13:45:38

Hello  > 1. as i know it, and correct me if i'm wrong, i have to calculate , (for > each conformation in my case), the difference in total energy energy > between the "solvated protein" (using solvent dielectric) and the > protein in vacuo. Actually, if you are simply interested in removing numerical "self energies", can just calculate the difference between the system with an inhomogeneous dielectric coefficient (solvent and solute dielectrics different) and the system with a homogeneous dieectric (everything set to the solute dielectric). If your solute dielectric is 1, then this corresponds to the calculation you are describing. > 1. for two grids with the same size but different spacing, i have > performed a solvation calculation (=totEsolvtotEvacuo), i received > different absolute results for total energy(totEsolv and totEvacuo), but > the solvation energies (=totEsolvtotEvacuo) stayed very close in both > grid resolutions. The total energies are not meaningful because they contain "self interactions"  the energy of charges interacting with themselves. In the limit of infinitesimal grid spacings, these selfinteraction energies should become infinite for the point charge model we're using. > in both cases i used Grid lengths: 58.761 x 32.129 x 25.828 > in which the protein fits very comfortably. > for a 161 x 97 x 97 grid: > totEsolv =9.534619136962E+04 kJ/mol > totEvacuo =9.875123989039E+04 kJ/mol > totEsolvtotEvacuo= 0.340504852 E+04 kJ/mol > for a 161 x 129 x 129 grid : > totEsolv = 6.803247450678E+04 kJ/mol > totEvacuo = 7.141364011794E+04 kJ/mol > totEsolvtotEvacuo=0.338116561 E+04 kJ/mol So you observed moreorless what I would have expected: convergence of solvation energies but strong changes in the absolute energies. > i read some posting about it, that said that the coulombic energy should > be added: > http://cholla.wustl.edu/pipermail/apbsusers/2003October/000104.html > can anyone give me a little deeper explanation about why it's done, and > how can this correction be gridresolutiondependant (since i need a > different correction for each grid obviously)? You should be able to get the total energy of your system by adding the Coulombic energy (in the solute interior dielectric coefficient) to the solvation energy you just calculated. This will put the interactions between charges back into the energy. > is there also an effect of > grid size (for a given spacing), because of the larget solvent volume? Possibly  this is usually due to a poor approximation in the boundary conditions. For example, the effect of an insufficiently large grid is more pronounced with "single DebyeHuckel sphere" boundary conditions than with "multiple DebyeHuckel sphere". > 2. what is a legitimate grid spacing (in angstroms) for calculations, in > order for the to be meaningful (but not necessarily very accurate)? Usually ~0.5 A. > 3. how can i estimate the error, in the energy calculations? (except for > running many calculations for different orientations on the molecule) > thanks a lot in advance, That's usually the best way. We have error estimate technology available with adaptive finite element methods but this is not implemented for the multigrid methods you are using. Thanks, Nathan  Assistant Professor, Dept. of Biochemistry and Molecular Biophysics Washington University in St. Louis http://cholla.wustl.edu 