Skip to main content

In Silico

Models and Reality

I have to admit that I enjoyed seeing this question asked: how come we still use wind tunnels in aerodynamic engineering? Why don’t we just model everything in software? The answers, from people who’ve actually done some of the work, are what you might expect: the models all involve degrees of approximation, gloss over some effects that are sometimes important, can help you but have to be given a reality check, etc. (Turbulent flow is no joke, theoretically or computationally, as any physicist can tell you).
The exact same sorts of answers, with a few nouns swapped out, could be given for the similar question of why we don’t just design drugs using computer simulations. I’m asked that often by people outside the field, and have run into many people over the years who assume that it’s just the way that drug discovery is done. But no, the continued existence of med-chem departments, such as they are, is testimony to the empirical nature of the business. It’s sometimes maddening, but reality can be that way.
Note: every good modeler I’ve worked with has made it very clear that they know that they’re working with approximations of reality – in fact, I think that’s a prerequisite for someone to be a good modeler.

18 comments on “Models and Reality”

  1. That’s one reason why I find the last sentence in this year’s Nobel Prize citation…well, manifestly disturbing (“Simulations are so realistic
    that they predict the outcome of traditional experiments”).
    This is probably also a good time to take another look at that paper asking why drugs can’t be designed like airplanes.

  2. annonie says:

    Maybe good modelers know about the approximations, but not all the med chemists do.

  3. simpl says:

    Surely all chemists are good modellers, imagining molecules like the ones in the magic methyls post?

  4. MolecularGeek says:

    Actually, that’s one of the two big things that we try to emphasize in our molecular modeling course. Most of the students taking it who enter industry will end up in early stage lead identification or optimization positions. By having them learn the basics of some of the software and do a small project related to their interests, hopefully they will learn how crucial good data from the bench is to the process, and what kind of results they can reasonably expect from a computational model. Yes, it would be cool if they learn how to use some of the new design tools in MOE or Schrodinger, but being able to ask hard questions of the modeling group where they end up is a very transferable skill.

  5. Anonymous says:

    They still use wind tunnels so that the models can prove they are wrong.

  6. H2L says:

    Software has provided immense value in the design of airplanes, but it cannot do everything. While the relative contribution of computation in drug discovery can be debated, conceptually it is the same situation. In short, computation should be used to complement experiments, and vice versa. I presume Derek and others who post here regularly believe this, but the discussions often turn into a “comp vs. med” competition. Maybe the limited successes of computation in drug discovery are in part a consequence of the two fields operating somewhat independently. Fortunately, it appears that things are changing with the new generation of scientists. Many comp chemists come from med chem backgrounds and more med chemists are using comp chem tools as a core part of their work. The move from comp chem to med chem is rare. That probably means something.

  7. Drug Developer says:

    George E. P. Box: “Essentially, all models are wrong, but some are useful.”

  8. Interested Outsider says:

    To all the scientists out there:
    Is there a distinction between theorists and modelers? Since one can argue that all our theories are essentially models, are the two terms synonyms?

  9. #8: That’s an excellent question. Models are derived from theories which are more comprehensive, rigorous and general. A model is usually a particularly instantiation of a theory applied to a particular problem. Here’s an example: Einstein’s general relativity is a theory, but its application to the problem of cosmology and specifically to the origin of the universe results in several models, the most widely accepted of which is the so-called Lambda-CDM model containing a cosmological constant and dark matter. Similarly in case of many molecular models, the underlying theories are those of quantum and statistical mechanics.

  10. dearieme says:

    An old acquaintance of mine said that “a model is a theory you don’t believe in.”

  11. Anonymous says:

    Every good looking model that I’ve asked out has suggested I get a better grasp on reality.

  12. annonie says:

    If you say it enough times, publish on it enough times, talk about it enought times, it almost certainly becomes “the truth”.

  13. Michael John says:

    There must be something to this modeling stuff.
    Compugen has reported ID’ing 9 Fc fusion proteins (validated in silico )from a single B7 discovery run. Total investment 6 months and less than 1 million dollars by a company with less than 100 employee’s/
    Bayer has signed on for two so they have had a close look at the work.
    Charles Drake (John Hopkins) Iain McInnes (U of Glasgow) Antoni Ribas (UCLA) and Howard Soule (Prostate Cancer Foundation) have signed up as their pipeline program advisory board.

  14. Anonymous says:

    @8: A theory is an idea or hypothesis to explain and predict observable phenomena, whereas a model is a simulation (or simulator) based on such hypothesis. I think.

  15. RedFiona says:

    The Marussia F1 racing, back when they were still Virgin Marussia tried to model their aero-stuff on a computer instead of spending the money on borrowing someone’s windtunnel. Needless to say, it didn’t work.

  16. Oblarg says:

    As an undergrad mechanical engineering researcher whose work involved dabbling in CFD (and I likely will resume doing so once my schedule allows me to work again), no one who has ever had to work with numerical modeling of fluid flow would ask this question.
    Aerodynamics are frighteningly hard to model correctly; the range of scales on which you have important effects is simply too gargantuan. Direct numerical solution is pretty much impossible with current computers, and the general approach to handling turbulence is basically to insert phony viscosity terms (with varying degrees of fanciness in how you do it) to damp out all of the small-scale stuff so you don’t have to calculate it. This, obviously, leads to rather significant differences between simulated flow and real flow, and ultimately a lot of what you end up doing is playing around with your model parameters until what you get seems to match what you’re observing in your wind-tunnel. Clearly, you can’t remove the wind-tunnel from the equation – you’d be fumbling in the dark, so to speak, with absolutely no confidence in your calculated results.
    Additionally, while a good numerical solution in principle ought to be grid-independent, fluid dynamics are anything but; getting a mesh around your geometry that gives you acceptable results is even harder than picking a suitable computational approximation.
    So, in short, it’s not a simple black box where you put in your physical system and it spits out a realistic simulation. It’s an iterative, back-and-forth process with tweaking one of numerous parameters and checking how well your simulation matches the empirical data.

  17. jbosch says:

    A model is only good if you can support it with data.
    VLS or in silico stuff can provide guidance but without an experiment to validate the model it’s just pushing numbers around and meaningless.
    Don’t get me wrong here, we do in silico stuff as well, but only to provide us with ideas on what to follow-up experimentally. And I have to admit we have some decent successes recently which we submitted for publication.

  18. Boing boing says:

    Don’t cherry pick folks. While CFD might be backed up with experiments FEA is sufficient to build airliners and other large, physical objects. Hell, they build nukes by modeling and I think we all can agree that’s complex…
    If the models work, we use them. If not we don’t.

Comments are closed.