From: Kitzmiller, Nate (
Date: Sat Jul 22 2017 - 23:44:30 CDT

We are working on the charge optimization step for the parameterization of a SHP-2 inhibitor molecule. We are unable to get the QM/MM distances to align. The optimized charges converge and the objective function reaches a minimum. The QM/MM energy alignments are sort of ok (not as good as it should be), but the distances are awful. A “tail” of our last ChargeOpt.log file is shown below.

One complication of the molecule is that a chlorine and 2 hydrogens form a tight triangle in the geometry optimized molecule. (We checked it more than once). This affects the

We have tried a number of remedies.

0. We fixed the charges for atoms whose penalties from Paramchem were zero. These were typically ring carbons (and their hydrogens) not near heteroatoms. (This is standard practice, we think, rather than a remedy.)

1. We excluded QM water interaction files (Gaussian log files) for which there were errors or for which the optimal distance was too far away (several angstroms).

2. We likewise excluded QM water interactions for donor interactions if the charge and the acceptor interaction showed it was clearly an acceptor. (And vice versa.)

3. We refined QM water setup files (Gaussian input files) for a number of atoms.

a. In the geometry-optimized configuration, 3 atoms (two hydrogens and a chlorine) form a triangle, and the usual choice of moving the water “straight away” from the ring resulted in high energy clashes with the neighbors. We reoriented the water molecules to move in a different direction that seemed to represent the most rapid departure from all other atoms. These QM files resulted in realistic equilibrium distances with the types (donor or acceptor) aligning with our intuition.

b. One or two other atoms, one notably an oxygen, needed a different direction for the QM single-point energy calculations. Similar noises about distances and atom types.

4. In the charge optimization, we increased the weight for the distance to 3. Some distance differences (as seen in either the COLP or the log files) are zero, but many are still large – as much as 0.4 A (or more?).

5. We lastly tried excluded QM water interaction files for atoms whose charge was fixed (from CGenFF).

We are at a loss as to what to try next. Grateful for any guidance.

I can send pdb/psf, geometry opt files, etc if they would help.

We are new to this work, but have read the CGenFF paper, the fftk paper and worked the tutorials and followed the workflow. I suspect we are doing something that someone with experience would say, “We never do that”, or not doing something that others would say, “We always do that.”

With thanks,