Skip to Content

In Silico

Using the Same FEP Ruler

With free energy perturbation having its time in the calculational spotlight, thanks to Schrödinger and others, it seems worthwhile to link to this new paper. It’s a proposal for a common framework to analyze the results of such work. That’s needed, because (as far as I can tell, as a definite outsider) every group seems to have its own idea of how to do that. This situation makes it difficult-to-impossible to compare various approaches, so even if this isn’t the best possible set of benchmarking tools (I’m not qualified to say), just getting everyone to use the same ones would be a step up. (Thanks to Ash/Wavefunction on Twitter for pointing this one out).

14 comments on “Using the Same FEP Ruler”

  1. Fluorine Chemist says:

    Am I the only one who thought that this post will be about a ruler made out of FEP (Fluorinated Ethylene Propylene)and was surprised why anyone would want to make one?

  2. A Chemist says:

    FYI, one of the authors sits on Schrodinger’s SAB.

  3. Ed says:

    Alternatively, one could use a hugely faster, much easier and equally useful technique altogether (eg HYDE – google it).
    by the time FEP is a practial and useful technique that sits in medicinal chemists desktop PCs, the (small molecule) pharma industry will in all likelihood no longer exist.

  4. FYI says:

    nobody at schrodinger has a fucking clue what they are doing. if they did, they predict something and do an experiment. instead they are running calculations on known results, and being laughed at by the scientific community for it.

  5. Ed says:

    #4 – why not tell us how you really feel? 🙂

  6. RM says:

    Ed@3 – While I’m sure Hideto Takarai is a fine musician if you’re into Japanese Alt Rock ( – the first hit when I Google “HYDE”), I’m not sure if I’d trust his judgement on the free energy of binding for small molecules.
    The admonition to “google it” doesn’t work when what you’re Googling is short and not uniquely spelled. (Which is something to keep in mind for people coming up with company names and acronyms for techniques – if it’s a common word, your online search-ability plummets.)
    After trying several search variations to limit the results from “Jekyll and Hyde” and “HYDE protein powder”, I presume you’re referring to the “HYdrogen bond and DEhydration energies” method from BioSolveIT ( – thought I can’t be certain if this is what you meant. I also have no clue as to what benefits you believe this particular technique has that makes it worth pointing out in particular.

  7. Ed says:

    Yes, I meant HYDE scoring function, but the gist of the argument was that Schrodinger are doing everything in their (significant) power to re-frame the scientific debate about predicting small molecule binding affinity in a direction that best suits their purposes because their current docking and scoring software (Maestro and Glide) have been consistently trailing competitor products in multiple independent studies of docking and scoring performance (e.g. pubmed ID 25682361), and are also trailing in terms of software interfaces and usability.
    They thus rely on their marketing machine to push this crap – see the multiple recent journal adverts (sorry, meant “articles”) on the retrospective validation of their incredibly slow, unreliable and inaccessible FEP technology using hugely expensive hardware with supporting info only provided in their proprietary file formats.
    Whether one likes AutoDock, Gold, ChemScore, FlexX-HYDE or some other particular flavor of docker isn’t my point – rather that equally useful tools yielding equivalent levels of accuracy are available today which work in seconds rather than days, without the additional hassle of having to buy and configure 600 graphics cards to make a simple prediction of binding affinity!

  8. Anonymous says:

    @7 You are referring to one paper that is only did a scoring power test, I believe there is more to docking than predicting affinity. E.g. have a look at . (“Scoring functions with good docking/screening power do not necessarily have good scoring/ranking power, such as GlideScore-SP, and vice versa, such as X-ScoreHM. “)

  9. H2L says:

    Seems pretty clear with all of these papers about FEP that there has been a paradigm shift in the field. There are just too many papers from excellent labs and in top journals for FEP to be just another computational method. Sure, the authors of these papers likely overhype the applicability of the method, and surely it is not going to replace the need for intense chemistry efforts anytime soon, but it seems pretty clear to me that FEP can provide big value when used in the right project. Based on this blog and the general sentiment of med chemists, there will be a majority of people who want to put there heads in the sand. Then there will be the early adopters, who will likely boost the company productivity and gain appreciation all the way up the R&D organization. I want to be part of that, but based on discussions with my colleagues I have the sense that we are going to be laggers. While I do not know enough about comp chem to know exactly which flavor of FEP is best, I think I know enough about physics to see that FEP is very different than docking, pharmacophore modeling, QSAR, and the rest of the tools that have not been transformative. I guess only time will really be able to tell, but I do not have enough time to wait and see.

  10. Anonymous says:

    @7: It is not realistic to put about docking programs and FEP in the same category. One is not better than the other in some abstract sense; they are just designed for different purposes. Docking programs are designed to predict poses and separate actives/inactives in virtual screening. That is it. And some docking programs can do that pretty well (and much better in the hands of someone who knows how to use them). FEP, on the other hand, is designed to predict binding affinity. One first needs a ligand pose, which could come from a crystal structure or docking, but then FEP (or other free energy methods) take over to predict affinity. Unless one does not believe in physics and thermodynamics, it is hard to imagine how a docking program could be better than FEP at predicting binding affinity. The key reasons why FEP lagged for so long are:
    1. Poor force fields
    2. Insufficient sapling
    3. Hard to setup/run/analyze
    It appears all 3 of the above have now been addressed to a large extent, and I am guessing that Schrodinger is not the only player in this game. I do not think there is any magic here (unless you consider physics to be magic). On the other hand, getting a docking program to accurately predict binding free energies would be more like magic, because they do not contain all of the necessary physics to accurately model that problem. We all knew that FEP would eventually be accurate enough to impact drug discovery. The time had to come (unless one does not believe in physics and thermodynamics). Someone just had to show that it works on a large enough scale on relevant systems. So, Schrodinger did that. Now all of the others will put the effort into implementing their own version, and I expect FEP will be used prolifically throughout pharma within a few years (except for the dinosaur companies that are unwilling or unable to do anything new). Let the free energy games begin!!!

  11. Ed says:

    The Schrodinger marketing machine swings into action! FEP is saved!

  12. Anonymous says:

    @7: Thanks for pointing out that the input structure used in the Schrödinger JACS paper ( ) were provided only in their proprietary file format. I requested that they provide the structures in standard file format and they have done so. The files can be downloaded here:

  13. Wavefunction says:

    #11 and #4: There are some clearly prospective results in that recent JACS paper by Schrodinger that Derek blogged about. Do you have specific complaints about these?

  14. wavefunction is a tool says:

    #13 no, those results are not prospective. every single experimental result was published before the calculations were attempted. if schrodinger knew anything, they’d publish the calculations first then do the experiments, but they don’t. friesner is an idiot and you’re his tool, wavefunction.
    we’ll all be better off with this company gone.
    last time i reviewed a schrodinger paper, the editor also worked for schrodinger but didn’t recuse himself. when i pointed this out they ignored my review and published the total garbage over my objections.
    ultimate proof in peer reviewed journals that friesner is either a moron or a scumbag (or both): he “forgets” to mention his ownership in schrodinger when publishing his glamour trash in PNAS (worst journal ever BTW). not once but twice:
    just hang it up, richy rich.

Comments are closed.