I would like to calculate the interactions between a couple of small ligand molecules with proteins, so I firstly used CGenFF to generate topologies for the small molecule, which are expected to be compatible with the charmm FFs in GROMACS. But the penalty obtained is too much, for example, “param penalty= 25.000 ; charge penalty= 45.141” for a protonated dopamine.
I assume this is caused by inaccurate atom charges? So is it correct to directly replace them with (restrained electrostatic potential charges) RESP or RESP2 charges got from Gaussian (surely after geometry optimization)? Would it make the simulation results more accurate? Also should I change other parameters in the topologies accordingly? Super thanks!
ref: Non-bonded force field model with advanced restrained electrostatic potential charges (RESP2)
Charges taken directly from QM are unsuitable for CGenFF as they will not incorporate an effective polarization response. Consult the CGenFF literature for very detailed descriptions of how charges are refined in the CHARMM force field (and thus CGenFF, as a component of it).
Many thanks for your kind reply! Are there any ways to lower the penalty then? I am very new to this topic. I have optimized the molecule geometry in Gaussian before, but don’t really know what else I can do.
I did have found out another way where the Amber FFs instead of Charmm are used and the topology of small molecules are generated by acpype where RESP charges are applicable. But some properties of the individual protein (without ligand) calculated based Amber FFs are different from the previous published results calculated based on Charmm FFs. I would like to keep the results consistent, so the Charmm FFs seem my only choice for this molecule-protein system being investigated. Do you kindly have any comments on this case?
In CHARMM, charges are assigned to overestimate gas-phase dipole moments on the order of 15-20% and also to reproduce molecule-water dimer interaction energies and geometries. The protocol for doing those calculations and how scaling factors are applied in certain cases are described in the CGenFF JCC paper. There are a lot of QM calculations that are required to get this right.
As far as AMBER vs. CHARMM, that’s really your choice in terms of what model you believe is most appropriate for your system. If you are building off of previous results using CHARMM, it doesn’t make sense to me that you would switch to AMBER and then try to compare the outcomes.