Add a section on apo-state sampling challenges in ABFE

[Nithishwer]

Issue description:

ABFEs can overestimate binding affinities when the apo state is not adequately sampled on accessible simulation timescales. This limitation shows up in several related ways and could be discussed either as a standalone section or distributed across the existing water-network, conformational sampling, and cofactor sections.

The main cases to cover are:

Water resolvation in occluded pockets: In some systems, water does not re-enter the binding site during non-interacting alchemical windows, which can bias ABFEs toward overestimation. A possible remedy is to introduce GCMC with alchemical windows as a equilibration step before production.

Cholesterol/lipid binding in membrane proteins: Some membrane proteins bind cholesterol or other lipids in the apo state at lipid-exposed pockets, but this can be difficult to sample because lipid relaxation is slow and expensive. Coarse-grained MD could be used to probe cholesterol residence times, followed by a separate FEP leg to account for the apo-state contribution.

Ion binding in the apo state: Similar to lipids, ions can stabilize apo conformations or occupy relevant binding-site positions, and may require an additional sampling strategy or FEP correction.

Conformational rearrangement of the pocket or protein: This is likely the most challenging and common source of overestimation. Apo and holo states may differ significantly in binding-site geometry or protein conformation. Ensemble analysis and suitable collective variables, such as those used by Vogele et al., can help diagnose these differences. Once apo and holo states are identified, possible corrections include seeding later ABFE windows with apo-like structures or using umbrella-sampling-like methods to map the free-energy surface between apo and holo states.

This section would frame apo-state accessibility as a general limitation in ABFE workflows and summarize practical approaches for diagnosing and correcting it. The checklist in section 11 would be modified accordingly.

[fjclark]

Thanks very much for the comments @Nithishwer. Personally, I agree that apo <-> holo differences/ apo state sampling is hardly discussed and should be more directly addressed (I can only see a brief mention of this when discussing water, and a currently commented out NEQ switching comment "Another example would be using true apo and holo protein structures as the end states to initialize simulations for absolute binding free energy calculations." ).

I think the main things to consider are:

a) To what extent are these unsettled practices (so should be added to the section at the end)? Maybe additions should be split between a brief new section in the main text and a paragraph in the unsettled practices section. For example, I think things like "Coarse-grained MD could be used to probe cholesterol residence times" and also probably "seeding later ABFE windows with apo-like structures" fall outside of the scope of currently-established best practices and would likely be more appropriate for the unsettled practices section. Maybe also worth mentioning that a common approach is to avoid all apo sampling challenges and just offset the results.
b) How do we avoid overlapping too much with what's already there, e.g. we already have "In absolute binding free energy calculations, a ligand is fully decoupled from a binding site cavity, which may otherwise have a different hydration pattern in the \textit{apo} state. In this case, the sampling of water molecules must be sufficient..." on line 1007)? Probably we can do this by being careful to cross-reference rather than rewrite and making some minor additions to other sections if required.

Does anyone else have any thoughts? @ppxasjsm @IAlibay @slochower @vgapsys @jmichel80.

What do you think @Nithishwer?

Thanks!

[IAlibay]

To what extent are these unsettled practices (so should be added to the section at the end)?

I generally agree with this. Outside of NEQ simulations, it's unclear that the maths and consequences of some of these suggested remedial approaches have been fully scoped out. Some very rough thoughts:

1. What happens when you don't sufficiently mix major state changes sufficiently (be it conformation or the presence of a water, etc...). How do you identify and handle such discontinuities?
2. Water equilibration is a good idea - but doing GCMC solely during equilibration (e.g. varying N between windows), can be problematic without care / accounting for it in the reduced potential. One approach is to rely on water swapping instead (keep N the same), but that's also not something I'd consider "settled" practice. I.e. do you have sufficient overlap in phase space that samples the changes in water presence? In most cases, does it truly provide better results than just equilibration at one state and using that solvation for all the windows?
3. Combining results from CG with atomistic MD has its caveats - knowing when the two are unlikely to agree is likely an unknown.

This is to say that having a section which mentions that there are potentially sampling issues with the things highlighted here seems reasonable. However asserting standard practice would require care.

making some minor additions to other sections if required.

In the spirit of not blocking a new release of the best practices, I would suggest very very minimal changes to the text where possible. Maybe a couple of extra sentences to acknowledge the issue and that knowing how to fix them is subject to ongoing research? (unless folks feel like these are settled questions)

[fjclark]

Thanks @IAlibay. In terms of timeline, at the earliest it will be a few weeks until we're ready for release, as there is a conflict we're in the process of resolving and Michael Shirts is planning some overall restructuring (I will update everyone on timing soon).

[Nithishwer]

I agree with what @fjclark and @IAlibay are proposing. I think unsettled practices would be the best way of presenting these. The idea is to warn the users of possible manifestations of the apo-holo issue and signpost them to literature that has dealt with these issues.

Maybe additions should be split between a brief new section in the main text and a paragraph in the unsettled practices section

I think this would be ideal

Maybe also worth mentioning that a common approach is to avoid all apo sampling challenges and just offset the results.

There is some literature to support the fact that offsetting retains ranking and can be cited.

Probably we can do this by being careful to cross-reference rather than rewrite and making some minor additions to other sections if required.

Could add a short paragraph and reference issues that are already mentioned.

Some thoughts on @IAlibay's thoughts:

What happens when you don't sufficiently mix major state changes sufficiently (be it conformation or the presence of a water, etc...). How do you identify and handle such discontinuities?

This is actually a problem we encounter with the seeding approach. It causes some compounds to be erroneously underestimated.

One approach is to rely on water swapping

Water swapping showed better results compared to using blanket solvation for all windows in one particular system. Although more thorough benchmarking is required before it can be a suggested best practice.

Combining results from CG with atomistic MD has its caveats - knowing when the two are unlikely to agree is likely an unknown.

Agreed. CG MD was just proposed to probe possible lipid/cholesterol binding to the pocket in case of a lipid exposed binding pocket.

I think all of these point towards inclusion of remedial methods in the unsettled practices (section 8) with a brief paragraph outlining possible issues in section 6.2

[fjclark]

Thanks @Nithishwer.

I think all of these point towards inclusion of remedial methods in the unsettled practices (section 8) with a brief paragraph outlining possible issues in section 6.2

Personally, I think this sounds good. It would probably be best to keep the section 8 additions concise too, in keeping with the other paragraphs there. @IAlibay are you happy for @Nithishwer to add these as tracked changes in the Overleaf for review? Thanks.

[fjclark]

Hi @IAlibay any thoughts on the above? Are you happy for @Nithishwer to suggest these changes in the Overleaf, and would you be happy to review? Thanks.

[IAlibay]

Sorry for the delay @Nithishwer & @fjclark

Folks mention section 6.2 and 8, however a lot of it also exists in section 5.2.

Here's my rough review of where we currently are at:

Water sampling

As mentioned, the issues regarding water sampling are discussed in 6.2.2. We already discuss the need to use enhanced sampling methodologies to deal with this in that section.

Lipid binding

This is somewhat discussed in 5.2 under "membrane proteins require additional care". Specifically we mention the challenges associated with displacing lipids from the binding site.

Ion binding

I'm not immediately seeing anything - this could be something to add somewhere in cofactors or conserved binding waters in section 5.2.

protein conformation re-arranagements

This isn't mentioned in 5.2 - it's something that is being hinted at around "the importance of configurational sampling" in 6.2 and "are there good starting structures in 5.2.

What I'm suggesting

My suggestion would be to maybe make some additions to 5.2 for ion binding and conformational re-arrangements and if needed small changes to 5.2 / 6.2 for the rest.

Then echoing @fjclark some concise additions on unsettled practices to address these things in section 8.

would you be happy to review?

I would be happy to.

[fjclark]

Thanks @IAlibay for the detailed comments. @Nithishwer could you please propose your changes? I'm going to be inconsistent, sorry, and go back on my previous suggestion: since this might involve small changes in quite a few places, could you please suggest changes as a pull request to this repo? This should make it easier for @IAlibay to review and avoid you having to point to all of the places in the Overleaf you've changed. The repo is up to date with the Overleaf. I can handle any merge conflicts once it's finished. Thanks.

[Nithishwer]

Thanks @fjclark and @IAlibay ! I'll the propose changes asap.

[fjclark]

Apologies for being inconsistent again: I'm away for the first 4 weeks of June and won't be able to handle PRs/ Overleaf synching then (which has become a bit fiddly due to Overleaf ownership issues). @Nithishwer, to avoid this taking a month to go in, please could you propose edits as tracked changes on Overleaf and be careful to point @IAlibay to them here when requesting review? Thanks.

[Nithishwer]

Hi @IAlibay and @fjclark,

I have now added some tracked changes in Overleaf addressing the apo-state sampling points discussed above.

• For apo-state solvation, I added a brief signposting sentence in Section 5.3, under the “Conserved binding site waters can play an important role in binding free energies” subsection, pointing readers to the unsettled practices section for ABFEs and water sampling.

• For lipids, I added a similar signposting sentence directing readers to the unsettled practices section

• For ions, I added a few lines in Section 5.3 under “Are there cofactors present?” to note that ions can modulate receptor behaviour and affect structural stability, which may in turn influence ABFE calculations.

• For conformational rearrangements, I added one sentence and signposted to the unsettled practices discussion in Section 6.5.

• I have added a concise paragraph to the unsettled practices section but please feel free to shorten it as necessary.

I have tried to keep the changes minimal and mostly cross-referential, in line with the discussion above. @IAlibay, please let me know whether these additions look reasonable. There is a couple of citations I need to add and will do it as soon as possible

[Nithishwer]

Hi @IAlibay Can you please have a look and let me know what you think?

Thanks.

[fjclark]

Thanks @Nithishwer!

@IAlibay could you please review? It would be great if we could get this closed within the next two weeks.

@IAlibay's review is the one that matters, but I couldn't resist adding a couple of comments.

For apo-state solvation, I added a brief signposting sentence in Section 5.3, under the “Conserved binding site waters can play an important role in binding free energies” subsection, pointing readers to the unsettled practices section for ABFEs and water sampling.

Sorry if I've missed it, but I can't see any tracked changes under this section, or a sentence matching the description above. Did you remove this?

For ions, I added a few lines in Section 5.3 under “Are there cofactors present?” to note that ions can modulate receptor behaviour and affect structural stability, which may in turn influence ABFE calculations.

A couple of comments on the added text:

Beyond their role as cofactors, ions can also act as allosteric modulators. In such cases, ions distal to the ligand may still be critical to include, as they stabilise particular structural states and a failure to model these accurately can prevent the protein conformational rearrangements necessary to sample the relevant apo and holo states. This is particularly important in the context of absolute binding free energy calculations as it requires accurate sampling of both the holo and apo states.

• Any references we should add with relevant examples?
• We might only care about getting accurate relative binding free energies from our absolute calculation, and may not mind having a shared offset due to failure to sample the apo state. Maybe clarify this (though you do discuss this in the unsettled practices section) or make the statement less strong?

I have added a concise paragraph to the unsettled practices section but please feel free to shorten it as necessary.

On the added:

A pragmatic alternative is to apply a reference-compound offset, this is reasonable only if the reorganization free energy is approximately uniform across the ligand series, an assumption that could break down with large conformational differences between binders.

Possibly worth adding several references to emphasise that this is the typical way this is dealt with, and it often works well in practice?

Thanks.

[IAlibay]

could you please review? It would be great if we could get this closed within the next two weeks.

Ah so sorry, yes will get this done asap (hopefully later today).