What we have learned

AMOEBA water has low diffusion constants. All the small molecules seem to have lower diffusion (we have a chart in small molecule manuscript). Although we have a second water model that has better diffusion AND better temperature dependence of density and diffusion(Figure 2 &13 in water-temperature paper). It would be interesting to learn what Teresa get from AMOEBA-v. We should encourage her to try.
I remember you have calculated water surface tension but I don’t remember what was the result. Higher surface tension will stabilize the protein more? Would AMOEBA-v give a different surface tension.
Protein unfold. Is this real? Is the RMSD within normal fluctuation (that correlates with local temperature factor)? If this is real, it means protein likes water too much. Dispersion or electrostatic?

We need to run more proteins!

Small molecule solvation free energy seems good (Vijay?). Meaning small molecule and water model are consistent.
Our trypsin-ligand binding free energy seems good. Unless there is fortuitous error cancellation, this indicates protein model is good. The ligand DMA used new gdma program. We should have relative binding energies of 2~3 more ligands in the next couple of months. What if we use new DMA on protein?
ion-water works from gas to condensed phase. Solvation free energy seems good. All free energies are good so far.

What we can possibly do to “improve” the model

Mostly the QM part. New DMA with high level basis set and new gdma (will be good to be consistent with nucleic acids now)
Backbone and side chain torsion. With really high basis set that is free of intramolecular BSSE. The tetrapeptide energy of Martin head-gordon with complete basis set does indicate LMP2 is over correcting BSSE for compact conformation, or over-favoring the extended structure.
Vdw? Unlikely
Valence terms? Pretty standard, except the bell torsion. I do remember the flexible bell torsion cause tetrapeptide to “slide” away from QM minima so I have to tune it down by adding regular torsions on top of it. Is David case seeing problem with amide groups?

Publish the model now, or never. We could get killed before we have a chance to respond to these attacks.
Run long simulations of a few proteins using TINKER, sander, PMEMD. I can help out if you need more CPUs. Based on your description of Case results, we should aim for 5-10 ns.
- Crambin
- Ubiquitin
- SUMO-2
- …
I will continue with ligand binding energy calculations, with more ligands and different proteins. It would be nice to have automated parameterization. My student is sometimes careless and I have to watch every step of parameterization.