Chris, it might appear that annihilation would suffer from the same exponential explosion you escaped from earlier. Is this avoided because you are still calculating the free energy difference between solvent with nothing and solvent with solute, rather than the absolute free energy of just solvent with solute? And perhaps there is some intuition here as well, which I think you discuss in your book, which is that there’s a tremendous cancellation of error that comes from sampling the A and B states in the same solvent ensemble. That cancellation cannot occur if you don’t have a reference state.
@rajdeepray39492 жыл бұрын
Nice lecture Professor. I have a couple of specific questions related to the FEP workflow for protein-ligand binding. 1. Do these methods generally ignore the de-solvation at the protein binding pocket? 2. It can be understood that since the de- solvation would take place for all ligands binding to the pocket, the delta(delta) G won't be effected. But however, it has been observed that small changes in the ligand do define the displacement of trapped water molecules inside the pocket. Can FEP account for these water energetics at the binding pocket??
@ChemProfCramer2 жыл бұрын
As your 2 questions taken together imply, it would generally NOT be a good idea to try to "grow" a ligand from "nothing" into a binding pocket filled with water, as it would be almost impossible to get a reasonable set of windows along the FEP trajectory defined by lambda that could sample the expulsion of water molecules from the site. But, with respect to small changes in solvation associated with one ligand being perturbed to another, that poses a much less difficult problem, so long as the solvent shell doesn't change drastically, e.g., in a rather tight pocket where solvent movement in and out is kinetically slow, it might be hard to model a situation where a change in ligand meant the total number of waters IN the pocket changed by an integer value. At the stage where the probability of one water molecule being in the pocket or out of the pocket was equal, an MD algorithm would have a VERY had time properly sampling that situation, and even for MC it might be quite challenging depending on how moves are made.
@linezero90164 жыл бұрын
On minute 9:16 you say that we run the simulation on the ensamble for A, then we calculate the energy difference for moving the H atom to state B. When we star using lamda, do we still run the simulation on the ensamble for A? or we do it for whatever value of lambda we are at. If we do the latter, how do we run a simulation for an ensamble of lambda... say lamda =0.05. I am assuming that there will be .95 of proton A and .05 of proton B coexisting, so should we take .95 and .05 of their masses when running the simulation?
@user-cj3lz3uw3l3 жыл бұрын
Hello Prof. Carmer, at 6:40, "ln" should be there in the last expression, right?
@ChemProfCramer3 жыл бұрын
good catch, yes (it's there on the next slide, happily...)
@user-cj3lz3uw3l3 жыл бұрын
@@ChemProfCramer Thanks a lot!
@saleheennoman7 жыл бұрын
Another related question; when I have a small energy difference between two end states, I can incorporate FEP technique to get the free energy difference. But when I'm looking at two end states who are separated by >2.0 eV, the number of windows between those states becomes too large and hence the process becomes computationally expensive. For example, if I'm looking at adsorption/desorption process of ethylene glycol and I'm trying to calculate the free energy of solvation for that step, If I want to slowly vanish EG from water, it gives me enormous number of windows. Is there an alternative approach for calculating free energy of solvation for adsorption and desorption processes?
@saleheennoman7 жыл бұрын
This question might not be directly related to this lecture, but still about related to the FEP calculation procedure; I was wondering, when we generate a finite sized MM conformation for image A (lets say I have an ethanol molecule immersed in water and there's a slight change in ethanol molecule between image A and B, my MM part consists of water and QM consists of ethanol), how much further apart the snapshots of water has to be to conclusively say they are uncorrelated? Literature survey shows large variations in their own conclusions. While this 1982 paper (DOI: 10.1002/bbpc.19820860308) from Dr. Leyte mentions the correlation time for re-orientation of O-H vector at room temperature to be (1.71+-0.07)ps, Dr. Fang’s work from 2014 (DOI: 10.1021/jp5009785) mentions the auto-correlation time length to be 10ps at the same temperature. Can someone shed some light on this issue?
@ChemProfCramer7 жыл бұрын
Different re-orientational motions of solvent molecules (when they are anisotropic, like water, for example), tend to have different characteristic decay times for their autocorrelation functions. My understanding, from listening to my ultrafast friends, is that very quickly reorienting librational motions can speedily relax local solvation fields in response to a solute change, even though much longer times may be required for complete reorientation to full equilibrium. However, I don't claim any special expertise in this area.
@saleheennoman7 жыл бұрын
Thanks a lot Dr. Cramer!
@linezero90164 жыл бұрын
In 0:35, how do you keep the partition function unitless?
@amritendukoley4 жыл бұрын
Hello Dr. Cramer, How many atoms at most would you suggest to change (delete/add) at a time for a FEP calculation?
@ChemProfCramer4 жыл бұрын
I don't think there's a simple answer to that, to be honest. First, it's probably not atoms that one should really think about, but more functional groups, where the number of atoms in the latter is really dictated by the nature of the group. Second, it will also depend on the degree of perturbation associated with the change -- the stronger the interactions being introduced, the fewer that should be attempted at one time. Of course, this is very general guidance, and you'd likely want a real expert to consult on any given specific case.