Advice on CGenFF penalty scores for a large, DFT-optimised molecule

User discussions
1

Hello everyone,
I am new to molecular dynamics parameterisation and would appreciate some guidance.
I am working with a relatively large organic molecule (99 atoms) consisting of a crown ether, a long alkyl chain, and a tryptophan-like moiety. I first optimised its geometry in Gaussian 09 using the B3LYP/6-311++G(d,p) level of theory. My next step is to generate force-field parameters for MD simulations using the CGenFF server, so I converted the optimised structure into a MOL2 file and submitted it.

The server completed the parameter assignment, but the reported penalties are:

  • Parameter penalty: 20.000
  • Charge penalty: 21.673

My questions are:

  1. Are these penalty scores acceptable for running MD simulations, or should I reparameterise the molecule before proceeding?
  2. If these penalties are considered too high, what is the recommended workflow to improve the parameters?
  3. Since I am new to force-field parameterisation, could someone recommend a practical step-by-step approach or useful tutorials for refining CGenFF parameters for a molecule of this size?

Any advice on how to interpret these penalty scores and whether they are acceptable for production simulations would be greatly appreciated.

Thank you,
Somdatta

2

The penalties indicate that basic checks should be performed (dipole moment and water interactions for charges, conformational energy scans for bonded terms). Note that CHARMM parametrization uses MP2/6-31+G* for geometry optimizations, not DFT. You should also simplify the molecule to only parametrize new parts; alkyl chains and indole rings are already well captured by the force field, so you only need to generate new parameters for other parts, then assemble your molecule from the constituent pieces.

3

Hi Dr. Lemkul,
Thank you again for your earlier reply. It helped me understand how to interpret the CGenFF penalties and that I should focus only on the chemically novel parts of my molecule rather than trying to reparameterize the entire 99-atom structure.
I also realized that the structure I originally submitted to ParamChem was not in the correct protonation state. In my first submission, the phosphate group was protonated (PO₄H₂), which gave an overall parameter penalty of 20.0 and a charge penalty of 21.673.
I have now corrected the molecule to the appropriate anionic phosphate form (net charge = -2) and submitted it again to ParamChem.
The new results are:

  • Parameter penalty: 10.000
  • Charge penalty: 10.836

From the generated .str file, it appears that almost all of the penalties are localized around the phosphate group and the connecting C–O–P linkage, while the alkyl chain, indole moiety, and the rest of the molecule have essentially zero or very low penalties. This seems to agree with your comment that only the chemically new region should be considered for further validation.

As someone who is new to force-field parameterization, I wanted to ask you whether, in your opinion, these new penalties are acceptable for further MD simulations, or whether you would still recommend refining the phosphate region (for example, by validating the charges and/or the relevant dihedral parameters using QM).

I would greatly appreciate your advice on whether these penalties are sufficiently low to proceed or whether additional parameter refinement would still be considered good practice.

Thank you very much for your time and guidance.

4

A penalty score of 10 is the threshold for what we consider to be reliable without additional validation, so it is wise to perform some basic checks as stated above.

5

Thank you so much for you guidance.