## VMD-L Mailing List

**From:** Mayne, Christopher G (*cmayne2_at_illinois.edu*)

**Date:** Fri Jun 26 2015 - 10:51:22 CDT

**Next message:**Francesco Pietra: "Removing all non polar hydrogens"**Previous message:**CPMAS Chen: "Re: vmd error xMDFF"**Messages sorted by:**[ date ] [ thread ] [ subject ] [ author ] [ attachment ]

Mohan,

43 atoms can be quite a large molecule to work on, depending on if you're including hydrogens in that count. For reference, I try to keep my fragments to 20 non-hydrogen atoms or less.

I have a couple of comments with regard to the objective function that are critical to understand. Please see the Methods and Algorithms section of our paper to better understand how the objective function works (Mayne et al J. Comp. Chem. 2013, 34, 2757–2770.)

A) The objective function is NOT normalized with respect to the amount of target data that is used for the optimization. Each interaction specified from the target data will contribute to the objective function; therefore, optimizations that include many interactions (i.e., more target data) will likely yield larger objective values. This means that one should usually not compare absolute values for the objective function between optimizations. The description for this term is given in equation (3) of our manuscript.

B) An add-on to the above comment is with regards to weighting factors. There are two places in which weighting factors can be used to tune the optimization 1) for each interaction specified by the target data (QM Target Data section; see equation (3), specifically noting the wi term ), and 2) differentially weight the contributions between the interaction energy, optimized interaction distance, and dipole moment (Advanced Settings section; see equations (3) and (4), specifically noting the wdist and wdip terms. "wint" is implicitly set to 1.0). These weighting factors are multiplied against the specified term which is then summed into the total objective function; care should be taken when comparing objective function values between optimization runs in which these parameters have been changed. For example, if the weighting factor for the contribution of the distance term is adjusted from 1.0 to 0.5, the objective value will necessarily go down, because the distance contributions are being reduced by 1/2. The resulting decrease in the objective function may not be indicative of a better optimization, just a lower contribution from the distance term.

C) The interaction energy and optimized distance terms represent summed contributions from the target data. The score from the dipole moment, however, is a molecular property which is computed from the single point energy calculation. To try and prevent the contribution of the dipole moment from becoming overwhelmed by a large amount of interaction-specific terms, the dipole term is scaled by the number of charges (see equation 4, specifically the Ncharges term).

At this point an obvious question arises: If comparisons in the objective function cannot be made between optimization runs which differ in weighting parameters or target data, what is the best way to judge the optimization success? This is where COLP comes in. I typically use the optimization function as a way of following optimization progress (i.e. does it converge lower than where it started); however, I judge the success of the optimization based on how well each interaction matches the QM data. This entails looking as the Energy and Distance data (the boxes on the left in the COLP window) for each interaction. The units here are kcal/mol and A for energy and distance, respectively, and represent the difference from the QM value. Here is where one should compare absolute values.

Hopefully this clears some things up.

Regards,

Christopher Mayne

On Jun 26, 2015, at 10:09 AM, Mohan maruthi sena wrote:

Dear all,

I am trying to generate parameters for a simple molecule consisting of 43 atoms. I have performed qm calculations for water interacting with molecule and visualized the data using COLP. I find that total, energy overlap but dipole moment and distance are very low compared to total. Could just provide some basic details regarding the optimum fit. How to make sure the charges obtained are reasonable? I am attaching the screeshot of the COLP image, I think this fitting is not correct? Could you please comment on this? I will be very thankful if you could elaborate on basic charge fitting procedure.

Thanks & Regards,

Mohan

<charge.png>

**Next message:**Francesco Pietra: "Removing all non polar hydrogens"**Previous message:**CPMAS Chen: "Re: vmd error xMDFF"**Messages sorted by:**[ date ] [ thread ] [ subject ] [ author ] [ attachment ]