CHARMM Development Project Forums User Discussion & Questions General Chemistry Discussions Interaction Energy and Binding

Hi,

I'm running MD simulations to calculate interaction energies between "carbohydrate_A" and a range of other carbohydrates and then use these energies as a metric for their binding. I want to rank the carbohydrates according to their binding affinity to "carbohydrate_A". This was recommended by scientists at a chemical computing software company.

However, since I started the simulations I've found a paper that mentions that the total potential energy of different models cannot be compared directly, only energies for different configurations of chemically identical models. Since I'm testing different carbohydrates, I will not have identical models. Would this be true if I use the command "INTE" for example? If so, can anyone suggest a better method to rank the carbohydrates?

Thanks!

Strictly speaking, binding is related to free energy, and not simply to the potential energy calculated by the INTEraction command in CHARMM. Unfortunately, calculating free energy for this system may be quite challenging; the role of solvent and the range of binding geometries or "poses" add considerable complexity.

You should definitely use interaction energies, not total potential energies. The caveats mentioned by Rick still apply though.

Thank you both for your responses! I just want to make sure that I'm understanding them correctly.

For the ranking of the carbohydrates, if I use interaction energies and not total potential energies, would I be comparing only the van der Waals and electrostatic contributions?

Thanks!

The INTEraction energy contains all energy terms involving some atom(s) in your first selection and some atom(s) in your second selection.

The interaction energy represents a subset of the total potential energy. In general, when the atom selections are for two separate molecules, the VDW and electrostatic terms between the two selections are the only terms which contribute.

Except for limited cases such as a deep, rigid binding pocket in a protein where the difference in entropic contribution may be minimal, the interaction energy alone may not correlate very well to binding.

Thanks to both of you for your prompt replies!

Rick, you mentioned that the interaction energy alone may not correlate very well to binding. Do you have any suggestions on what else I could use in combination with these energies in order to make "decent" binding predictions? I have very limited access to experimental data at this point. (From your previous post in this thread and from literature, I know that calculating free energies can be challenging and resource-intensive, that is why I'm calculating interaction energies.)

Thanks!

There is unfortunately no free lunch as far as free energies are concerned, and in any case it may take a long time to converge. The Linear Interaction Energy (LIE) method of Johan Åqvist, or some variation of MM-PBSA/GBSA may be useful. Is your main interest in quick scoring or in more accurate and physically interpretable numbers (when a proper free energy calculation/FEP is probably the best we can do)?

The main disadvantage of LIE is that one has to calibrate the scaling factors against experimental data specifically for the compounds being studied, which may be a problem given that
MM-PBSA/GBSA seems the most viable option here.

Edit: or, of course, MM-GBMV or MM-GBSW.
Or you could try FEP anyway, using CHARMM-GUI's new Ligand Binder feature. That should take some of the complexity out of it (but I don't know how stable and well-validated it is, and it's most probably still computationally expensive).

Thank you Lennart for your reply and suggestions.

Because the long term plan is to have experimental work to verify our simulations, I think a quick scoring will be sufficient. Would you suggest using interactions energies for quick scoring? Or even then we should use some variation of MM-PBSA/GBSA?

Thanks,