Dear Dr. Gapsys,
Thank you very much for the pmx tool, it looks very promising to use for my PhD research!
I would be very grateful if you could help me with two interconnected questions (and one auxiliary):
- For me it is a usual situation that I need to explore transformations like A->B and A->C, i.e. starting from the same initial state A. At the same time, I find it difficult to imagine how could I implement the pmx strategy to run the dynamics of state A only once.
For A->B transformation I usually run the dynamics of state A already with dummy atoms of state B in the structure and with their hybrid topology. But if I had, for example, just production of state A without the dummy coordinates of state B in each frame and without hybrid topology, could I still somehow use this production for non-equilibrium A->B transitions?
I was thinking of creating the hybrid topology for each frame of state A production with respect to the same A-B pairs file and ligand B .pdb structure. This should put the needed dummy atoms of state B in each of the frames before running alchemical step.
Could you please tell me if it sounds adequate to you?
- As I understand it, the equations of motion during dynamics are solved for dummies just as for any other atoms, and therefore when I start the dynamics of state A with hybrid topology, the positions of the dummy atoms will be adjusted to positions of the other atoms of the ligand A (dummies has 0 charge, but we still write all the force constants for them in topologies as I can see).
If so, is it fine if I start the short nonequilibrium switches (50 ps) from manually created dummy positions, as I suggest in my question 1), i.e. not after their positions were adjusted by dynamics as in classic pmx tutorials? Or maybe it is better to give to each frame some short time for re-equilibration?
- Auxiliary question about the sign of dG - it seems to me that if I do the A->B transformation, the energy analysis script of the pmx gives dG as G(A)-G(B), which is vice versa to “intuitive” approach in classical physical chemistry where dG = G(B)-G(A). Could you please confirm that I’m right/wrong? Sorry for somewhat too obvious question. I did not find this clarification anywhere explicitly in the pmx output, and have infered it just from common sense of some of pmx tutorials / my own results, therefore I’m not sure.
Dear Dr. Gapsys,
Thank you so much for this answer!
As to the sign, I’ll test myself on some benchmarks from your 2020 ChemSci article to ensure I’m doing everything correctly.
As to creating dummies from different frames - I think I got the idea, and will test it soon.
But in connection with this question of creating dummies for each frame - I can imagine situations where, if wanted to create a topology for each frame of state A ligand using some fixed geometry of state B ligand, I would get different pairs files and therefore different topologies - depending on which conformation of state A I have in particular frame, or even depending on some issues like too long bondlength (accidental, during dynamics) or different positions of hydrogens connected to one same atom in different frames of state A. So I would be grateful if you may help me with some questions in connection to this:
Do you think that it is theoretically appropriate to run final energy analysis using many A->B transition energies, clusters of which were obtained with different common topologies? Of course, the different topologies of frames caused by different conformations of state A will probably give us bimodal distributions anyways, which is as I understand undesirable. But here I’m asking whether the fact that two A->B energies were obtained using different common topologies makes it absolutely inappropriate to include them in the same BAR analysis. These could be minor differences between topologies, like having two hydrogens represented as dummies in one case and not having in another.
So, let’s say we have productions of state A and state B without common topologies. Let’s imagine that ligands in states A and B sometimes change their conformations. Have you ever thought/done smth like generating new pairs files/common topologies for each pair of frames, like frame_1_A/frame_1_B, frame_2_A/frame_2_B, and run alchemistry like that? As I understand, even if we get non-unimodal transition energy distributions, at least each transition could have the most adequate common topology for its particular case. Otherwise, our same-for-every-frame topology may not adequately describe conformations, on which it was not based.
You don’t need to rerun atom mapping for every frame: use the same identified pairs and just create hybrid structure.
No, topologies must be identical for all frames in a transition.
You shouldn’t generate new atom mappings for every frame.
Thank you very much for your reply!
I was just thinking of potential difficulties of nonequilibrium approach: for example, when state A frames have two conformations, and state B has one conformation. 1st conformation of A is very similar to state B, while 2nd is different. If I use the common topology between states A and B, based on 1st conformation of A, the dissimilarities of 2nd conformation of state A between state B are not taken into account. Thus, 50 ps transition may be not enough to obtain correct conformation of B when alchemically moving from 2nd conformation of A into B.
Probably, in such situation there is no way other then separate the analysis of these two conformations of state A and create two different common topologies between state A and state B. After your answer it became clear - I can not work with different topologies within one energy analysis. Thank you again!
For this scenario you might want to disable MCS based atom mapping and rely only on the geometry based alignment. Also, make sure that the two conformations in the stateA do not inter-convert.