Talk:Implicit solvation

A problem with electrostatic implicit solvation models

There is a problem with electrostatic models, such as PB, GB, or GBSA, when they are applied to proteins with ionizable residues. If a charged residue (Asp, for example) is transfred from water to a nonpolar media (such as membrane or protein interior), this would be eneretically very costly according to calculations with GB, for example. However in the real life, this residue will simply becomes uncharged, which costs very little, 2.3RT(pH-pK), as has actually been observed for protein mutants with buried ionizable residues and hydrophobic alpha-helical peptides in membranes with a single ionizable residue in the middle. If I understand correctly, such ionization effects are not included in PB, GB, or GBSA. As a result, the energies calculated with PB, GB, or GBSA will be grossly overestimated. Do you agree? And if you do, should this be mentioned? Biophys 22:04, 31 October 2006 (UTC)

This sounds very interesting. Do you have a paper reference that talks about this? Jorgenumata 09:44, 1 November 2006 (UTC)

Yes, there are references. I will add some text with the referenences later. Biophys 13:17, 1 November 2006 (UTC)

"Old-fashion" implicit solvation model

The initial version of the implicit solvation model was to use different solvation parameters for different types of atoms (not the "uniform" value of 40 cal/mol/A for all atoms including N or O), and include some electrostatic contributions only for ionizable groups. For example, such approach was implemented in FANTOM. I think this should be described. Biophys 01:10, 1 November 2006 (UTC)

This is addressed. Look at the section "Ad-hoc fast solvation models". It says that A first generation of fast implicit solvents is based on the calculation of a per-atom solvent accessible surface area. For each of group of atom types, a different parameter scales its contribution to solvation. It then goes on to explain the second generation, like Karplus Lazaridis EEF1. If you have specific information about FANTOM, please include it. Jorgenumata 09:43, 1 November 2006 (UTC)

Yes, I see. The text you have written is very good. I just would like to say more about solvation models in the Eisenberg-McLachlan style. Also, such models can be applied for different puproses, such as modeling of protein folding or modeling of protein-membrane interactions. So, more can be said about certain practical applications (one of them FANTOM), with the corresponding references. I will make some additions that you are welcome to edit (may be next week). Biophys 13:15, 1 November 2006 (UTC)

I also have a question. What reference says that Poisson-Boltzmann equation belongs to implicit solvation models? I am not sure this is right. Biophys 13:27, 1 November 2006 (UTC)

That's not all the PB equation can be used for, but it is definitely used as an implicit solvent model - see this paper for an example. Opabinia regalis 04:56, 2 November 2006 (UTC)

Thank you! I thought this equation might be represented by a separate Wikipedia article, since it is multi-purpose. But this is not really important. Biophys 23:41, 2 November 2006 (UTC)

Actually it also has its own article - Poisson-Boltzmann equation. It's pretty stubby right now though - it didn't even have the equation in it till recently. Opabinia regalis 01:12, 3 November 2006 (UTC)

Problems and limitations

I can try to write down something about limitations of the implicit solvation models. Such limitations are important to know.Biophys 01:10, 1 November 2006 (UTC)

That would be useful. I never use implicit solvent but I'm actually pretty impressed with the quality of the results some people report, considering how bad the approximation should be. Opabinia regalis 01:12, 3 November 2006 (UTC)

Good comment! This method may work well because derivation of atomic solvation parameters from experimental transfer free energies (say water-cyclohexane) of organic compounds is very straightforward. Therefore, implicit models can capture energetics of solvations (with all its hidden entropic contributions) much better than MM or MD with all their serious defficiencies (see what I have written about that in force fields) (so the "implicit solvation" is not to save the computational time, this is an alternative approach). However, there are two things to remember. First, these solvation parameters must be properly determined (taking ad hoc values of 20 or 40 cal/mol/A2 for all atoms including C, N,... is not a good parameterization; and the interior of proteins and membranes is not wet octanol). Second, such parameters should be applied only to describe solvation or transfer of atoms between uniform media. One can express van der Waals attractions in solid state in surface enegy units or volume energy units, but this approximation is too crude (it does not capture "specific" distance-dependent interactions which are responsible for clustering of groups with similar polarities in protein structures). Biophys 17:03, 4 November 2006 (UTC)

ASA section first?

Biophys, you've been adding some very useful and detailed information here lately. I wonder, though, if it makes sense to put the current section on accessible surface area first? If we assume that readers either aren't familiar with the subject or are only vaguely acquainted with it, this seems like rather detailed information for the very beginning of the article - particularly the subject of flaws in parameterization should probably come after the sections about the models themselves. Thoughts? Opabinia regalis 04:40, 16 November 2006 (UTC)

I think ASA section should be first because it was historically the first, and the electrostatic models (except PB equation which is more general) are a kind of addition to the simplest ASA-based models. But maybe you are right that some of the things in the first ASA section actually belong to "Problems" and could be moved there. That would simplify the initial ASA section. Let me think about it. Maybe I can fix this. Biophys 16:17, 16 November 2006 (UTC)

Yes, I will correct this. Only first paragraph should remain in the introductory ASA section. Biophys 16:32, 16 November 2006 (UTC)

Great, that's much clearer. I did some minor writing tweaks and organization to the problems section, convertinig from a list to prose in subsections - it's easier to navigate when distinct segments appear in the table of contents.

If you don't mind, can you take a look at Accessible Surface Area vs solvent excluded surface? I've just proposed a merger of the two articles, as it's extremely silly to me that accessible surface area in lower case redirects to solvent excluded surface, but there may be a reason I'm not thinking of to keep them separate, and I'm not sure which of the three is the best title. Opabinia regalis 05:33, 17 November 2006 (UTC)

I agree about the merger. But we should not lose any important content from the articles during the merger. I think best name would be accessible surface area. It is most frequently used in the literature. Biophys 00:42, 18 November 2006 (UTC)

Great, merged to accessible surface area. Thought I should make sure I wasn't losing my mind about why there were two of them :) Opabinia regalis 04:24, 18 November 2006 (UTC)

Looks good to me. I also like the links. It is a shame that authors of Getarea do not provide its code as a standalone program. I asked but get nothing, and it would take too much time to dig out the code of Getarea from FANTOM (Getarea is unique, since it provides fast analytical calculations of ASA partial derivatives with respect to atomic ccordinates). Biophys 04:45, 18 November 2006 (UTC)

Point about sampling

Since I'm not really being clear on the sampling issue and won't be around much for the next couple of days to discuss it, I'll just quickly expand on the point here, which is essentially the point made in the Zhou paper. Implicit-solvent simulations aren't just a way of more quickly and broadly sampling the same distribution of conformations as explicit-solvent simulations; using implicit solvent (usually) gives a different free energy landscape with different minima and different transitions, which may not correspond to a) those from explicit solvent, or b) experimental native states. This is closely bound enough to the parameterization problem (especially overstabilizing salt bridges) that it can be considered the same issue, but I've run into enough people (experimentalists dabbling with simulation, mostly) who think implicit solvent is 'a faster way of getting the same results' that I think this should be explicitly stated. More generally, I think the article should be clear that implicit solvent is an approximation - if you had infinite computing power you wouldn't use it, except possibly in very limited special cases. Opabinia regalis 20:36, 22 November 2006 (UTC)

Of course this model is an approximation, so it definitely can be defined as such in this Article. Unfortunately, all force fields in Chemistry are also an approximation. Which approximation is better? That depends on what exactly you would like to model. The advantage of ASA-based models is that they operate directly with solvation free energy, not with enthalpy in vacuum (to simplify) as MM or MD, as I said above in this Discussion. The advantage is not the computational speed or bypassing the sampling problem (although this is true). Let's imagine that you have an unlimited computational power (Blue Gene Project). Could you fold the protein using MM or MD? Certainly not - for the reasons explained in force field article. People are even unable to predict the structural effects of replacing a single amino acid residue in a protein (if they could, the homology modeling problem would be solved). That is why people started using ASA-based models and a lot of other new empirical scoring functions, although this is usually not stated clearly in their papers for the reasons of political correctness (the paper by Levitt in Nature Structural Biology is one of few exceptions). If this is not clear, maybe we have to re-phrase slightly this article.Biophys 21:08, 22 November 2006 (UTC)

I think your recent edits make the subject clearer. Maybe it's my bias (all-atom purist :) or this isn't as common a misunderstanding as it seems to me, but it appears that there's some confusion among people not already familiar with the methods about why everyone doesn't use implicit solvent because 'isn't it the same?'.

While current force fields (and water models) clearly suck for a wide variety of reasons, you might be a tad too pessimistic... you have to resort to a bit of trickery, but Pande's group did, plausibly, fold villin. I'm not convinced yet that the current generation won't be able to fold even a protein with a very well-defined native state, since there are so few trajectories run into experimental folding timescales. Solving the homology modeling problem sounds like a tall order, since predicting the effects of a single point mutation (often done at a very qualitative level) doesn't directly translate into predicting the effects of changing 70%+ of the residues, adding or changing the lengths of loops, etc. Opabinia regalis 07:49, 23 November 2006 (UTC)

Question for the Experts

If the generalized Born model "does not include entropic effects" then why is it stated as ΔG and not an enthalpy? Ggrieves 17:54, 25 July 2007 (UTC)

Good question! The original Poisson-Boltzmann equation icludes only electrostatic energy (a part of enthalpic component). However "Gs" appears in the next equation (Generalized Born). This is because people who are doing this trick often claim to calculate free energy, although actually they do not. In addition, they are trying to combine electrostatic energy with empirical free energy of solvation calculated using first equation (deltaASA formula). I personally think that PB and GB models are only good to create an impression of "good science" and prefer the simplest ASA-model, because it actually works if properly parameterized (within certain limits as everything). If you want to improve this article, you are very welcome.Biophys 20:07, 25 July 2007 (UTC)

PB and GB model the electrostatic free energy, which includes an entropic component. However, this is not the whole configurational entropy of the water. PB and GB are mean-field approaches which do not account for directional interactions with the solute that alter configurational freedom. At room temperature, the hydrophobic effect is mostly due to a change in this kind of entropy, and is thus not included. It has been known for many decades that the hydrophobic effect is the most important driving force for protein folding. It is thus a strange state of affairs that the modeling community has devoted so much effort to electrostatics and neglected entropy and the hydrophobic effect. I agree with Biophys that PB is sometimes abused. It is used to give the impression that a complete free energy was calculated, when in fact only part of it was estimated. Jorgenumata 11:01, 11 October 2007 (UTC)

Rename?

To Implicit solvent models ? Biophys (talk) 13:45, 12 June 2011 (UTC)

Sounds better. • Jesse V.^(talk) 06:29, 20 August 2012 (UTC)