I have been trying parameterize a small part of a bigger molecule(91 atoms), which have a charge penalty of 25 and dihedral penalty of 45.
I understand that these relatively high penalties indicate that the corresponding parameters should be validated/re-optimized rather than used blindly.
I initially tried to optimize the parameters using the fftk+ORCA workflow at the MP2 level. However, I am having difficulty obtaining good convergence between the QM and MM energies.
I went through the charmm36 parameter files and found that the dihedral I have a high penalty is available in the carbohydrate parameter file(a similar chemical environment). Would it be reasonable to use the existing CHARMM36 carbohydrate dihedral parameter as the starting value for optimization/validation, provided that the local chemical environment is sufficiently similar?
There is also another issue. When I use the CGenFF-generated topology directly in GROMACS, I obtain extremely large potential-energy values for the molecule.
Would ffParam be a better option for parameter optimization in this case, or would you recommend continuing with the ffTK + ORCA workflow?
I would appreciate any kind of help and advice on how to tackle the parameterization and validation.
Parameterization-related questions are always extremely tricky because it can depend heavily on the molecule of interest, and without further insight only general recommendations can be given. First of all, it is good that you are sceptical of the parameters given by CGenFF and not trusting them blindly! :)
Which parameters are not converging that you are trying to refine, i.e., partial charges or bonded interactions? Usually you start with partial charges, then bonds/angles, followed by dihedrals. If the problems are with dihedrals, you can try to “play around” with the multiplicity/number of summands; however, this can be tricky and prone to overfitting!
If there is a similar dihedral existing in the CHARMM36 carbohydrate force field, you can use it for further refinement, but you should first check how it performs — maybe it is already good enough for your molecule.
Pure potential energies are not really meaningful in the CHARMM force field, from my understanding.
Regarding validation: in the best case, there is some basic thermodynamic data available for your molecule (enthalpy of vaporisation, solvation free energy) that can be calculated from MD simulations with your optimised parameters. Please check ffTK or the original CGenFF paper for more insight.
Regarding your question about (semi-)automatic workflows (ffparam, fftk, etc.) for parameterization: I have used both, and both have their (dis)advantages, like probably all other scientific software out there. There is also a new ffparam (v2) available that I haven’t used yet — you probably need to write a short request to the authors, since it is not yet available from their website. If you have access to Gaussian, you could have a look at GitHub - mackerell-lab/cgenff5: Data and scripts for CGenFF 5.0 paper · GitHub — they also provide some tools for parameterisation.