Good Math, Bad Math

In my post yesterday, I briefly mentioned the problem with simulations
as a replacement for animal testing. But I’ve gotten a couple of self-righteous
emails from people criticizing that: they’ve all argued that given the
quantity of computational resources available to us today, of course
we can do all of our research using simulations. I’ll quote a typical example
from the one person who actually posted a comment along these lines:

This doesn’t in any way reduce the importance of informing people about
the alternatives to animal testing. It strikes me as odd that the author of
the blogpost is a computer scientist, yet seems uninformed about the fact,
that the most interesting alternatives to animal testing are coming from that
field. Simulation of very complex systems is around the corner, especially
since computing power is becoming cheaper all the time.

That said, I also do think it’s OK to voice opposition to animal testing,
because there *are* alternatives. People who ignore the alternatives seem to
have other issues going on, for example a sort of pleasure at the idea of
power over others – also nonhumans.

I’ll briefly comment on the obnoxious self-righteousness of this idiot.
They started off their comment with the suggestion that the people who are
harassing Dr. Ringach’s children aren’t really animal rights
protestors; they’re people paid by opponents of the AR movement in order to
discredit it. And then goes on to claim that anyone who doesn’t see the
obvious alternatives to animal testing really do it because they
get their rocks off torturing poor defenseless animals.

Dumbass.

Anyway: my actual argument is below the fold.

So – what’s the problem with simulation? It’s not that we don’t have
enough computational power. The amount of computational power available to us
is truly mind-boggling. Many things that would have been completely impossible
just a few years ago have become absolutely routine. Sitting on my lap, I’ve
got a computer with two CPUs, each running at 2.5 gigahertz. Under my desk,
I’ve got a computer with 8 CPUs. And I routinely run programs that use
several thousand CPUs in a datacenter. That’s not unusual
today: people who want to run simulations can easily and cheaply
get access to clusters of thousands of computers.

So computer power isn’t a problem. Using modern computers
and software, we’ve easily got enough power to run incredibly complex
simulations.

And that doesn’t help with using simulations to replace animal testing
at all. The problem with using simulations for testing has absolutely
nothing to do with computational power. No matter how much computational
capability you have available, there’s still a huge, fundamental problem with
simulations. It’s not particularly hard to understand: simulations only do
what you tell them to.

Let’s start at the beginning: just what is a simulation? It’s a
model of a real system, which attempts to reproduce the effects
and/or behavior of the system that it models. In the case of computer
simulations, which is really what we’re talking about, we produce a
mathematical model and algorithmically describe how the model evolves over
time. In other words, we write a computer program that runs a mathematical
model of a process.

And right there is the problem. We produce the model, and
we implement the model. It does exactly what we tell it to. We
programmers have a saying which applies particularly well to simulations:
garbage in, garbage out. Computers do exactly what you tell them to; if what
you tell them to do isn’t right, then no amount of computer power is going
to change the fact that you didn’t tell them to do the right thing.

If we really understand what we’re simulating, we can do simulations that
are astonishingly accurate. For example, we can do wind tunnel simulations of
aircraft designs, and the results of those simulated experiments are perfect
to within our ability to measure them. We’ve got it down well enough that we
can get more precise results with a computational model than we can
with a scale model in a wind tunnel. We understand how air flow works. We
know how to model it. It takes a whole lot of computation to do a decent job
of it, but as I said before, lack of computational resources isn’t generally
a problem.

But we can’t simulate something if we don’t know how it works. And that’s
the problem with biological simulations: we don’t know how the
biological systems work. If we don’t know how they work, we can’t build
an accurate model. And if we don’t have an accurate model, we can’t build
an accurate simulation.

When it comes to biology, we just don’t know enough to
be able to accurately or even meaningfully simulate many simple
processes.

Let me give you an example. In 1956, a scientist named Otto Warburg
discovered that most cancer cells have an abnormal metabolism. Normal cells
metabolize sugars to produce energy using their mitochondria. Cancer cells
don’t usually use their mitochondria – they use an entirely different
metabolic pathway. It’s called the Warburg effect. (My
friend and blog-father Orac sent me a note to say that the Warburg effect was
actually discovered in 1906 – fifty years earlier than the citation I found would suggest. So instead of having 50 years to study it as I write below, it’s over a hundred!)

We’ve known about the Warburg effect for over fifty years. And
yet, we still don’t know why cancer cells do it. We don’t
know what causes it. There’s no way that we can simulate it. We can’t
write a simulation of a cancer cell, because we don’t have a sufficient
understanding of its metabolism: we don’t know when it’s going to use
Warburg. And the metabolism is one of the simpler aspects of
cancer behavior: things like how the DNA in the cell changes in
cancer is vastly more complicated. We’re just no where
close to understanding it – which means that we’re
no where close to being able to simulate it.

So we can’t simulate it. Or, to be more precise, we can’t
produce a simulation that we know is valid. We can’t simulate a process
that we don’t understand. We can’t simulate a process that we don’t
know about. And biology and medicine are just chock full of processes
that we don’t understand, or that we aren’t ever aware are occurring. So
we can’t simulate those.

There’s another aspect of simulation that’s important: validation. Even in
the best simulations, you can’t be sure that the simulation is correct until
you test it. That testing is called validation. To validate a
simulation, what you need to do is to take some starting point in both the
real world and in the simulation, and observe both for some period of time,
and then check that the results of the simulation and the real-world match.
For example, for air-flow simulations, you put an object into a wind-tunnel and
measure everything; then you simulate putting an object into a wind-tunnel
with the program; and then you campare the results.

Without validation, you have no way of knowing if your model is
correct.

Validation is really key when you don’t understand what’s going on. I’ll
pull out another example. I come from a family with a history of clinical
depression. I suffer from it myself. For me, and many other people, there’s a
class of drugs called selective serotonin reuptake inhibitors (SSRIs). SSRIs
were developed based on a theory about what caused depression – specifically,
that depression was caused by a shortage of the neurotransmitter serotonin in
the brain. So they developed a set of highly targeted drugs that muck with the
serotonin chemistry of the brain. And lo and behold, it works. It’s
often cited as an example of the amazing success of modern targeted
pharmaceutical development.

Except that all of the most recent research about depression appears to
show that the serotonin theory of depression is wrong. And yet, for a
significant number of people, the drugs work – and they work dramatically
better than you’d expect from placebo effects. They do increase the
availability of serotonin in the brain. They also do a bunch of other things -
like stimulate cell growth in certain parts of the brain. How do we figure out
how SSRIs work? Among other methods, computer simulation with validation from
animal models. We start with a model: what would we expect to see if the
serotonin model were correct? We work out that model in detail. We convert
it to a program that predicts what kinds of biological effects we should
expect to see if it’s correct. Then we do the test on a group of animals,
and check to see if what we observe in the animals matches what we expect
from out model. Based on the results of that, we can judge how well the model
matches reality.

From all of this, it might sound like computer models are boring and not
terribly useful. After all, all that they can do is what we tell them to. So
don’t we always know the results before we start?

Fascinatingly, no. We have some amazingly precise models of physical
phenomena, but understanding what those models mean when they’re scaled up to
life size is incredibly difficult. And in fact, sometimes, we simply don’t
know how to see what a model says about how a system will evolve through time
without actually watching it progress. For example, I mentioned computational
fluid flow above. We don’t know if it’s possible to compute the
result of a fluid flow system without simulating it through discrete time
steps. Simulation is incredibly useful, because it lets us take dynamical
systems, and watch how they evolve over time.

So simulation absolutely has a role. And hopefully, it can reduce
the number of animals that are used in medical experimentation. But it can
never replace them. There really is no substitute for reality.

Comments

  1. #1 Flaky
    March 1, 2010

    I wouldn’t say that computing power isn’t a problem. Suppose that a simulation of a human body, sufficiently detailed to replace animal testing, requires simulating all physical interactions at the atomic level – we don’t know if such a fine level is needed, but we can’t really rule it out – in that case there just isn’t enough computing power in the world to do that kind of simulations. And there isn’t going to be in quite a while.

    Of course for a quick refutal, one can just ask the animal activists for a solution to the protein folding problem, since that must by necessity be solved for computer simulations to replace drug testing on animals.

  2. #2 mdb
    March 1, 2010

    You leave out an important point, even if the pathway is known and well studied, it is impossible to know how a given drug or treatment will interact with an entire complex system. Say they develop a treatment to reverse the Warburg effect, it works great in the computer models – would you take it? No way – you’d have to be a moron. Your body, even with cancer, does many, many, many other things. Things that are unknown, things that are not understood, things that are known and understood, but are WRONG and finally things that are known, understood and right. A model can only account for those variables that are known and understood (easily less than 50% – medicine is no where close to an exact science). Of that subset, many will prove to be wrong (maybe 5% – think how many times medical advices changes). Betting your life on less than 50% chance that a model is right is stupid. I would put the odds that it was right much lower (20%).

  3. #3 jerthebarbarian
    March 1, 2010

    Thank you. The “just use simulations” argument is one of the dumber ones that comes out from animal activists.

    I think I understand where it comes from – there’s a compelling argument to be made that we no longer need High School Biology students – or even freshman-level college students – to do animal dissections anymore. At the level that they’re working at, a simulation is possibly at this point more than adequate enough to provide educational instruction for the concepts being taught. And that’s where I first heard the “use a simulation instead of real animals” arguments – in the context of intro students doing animal dissections to learn about biology.

    But it’s an argument that doesn’t scale up to research or even to higher level instructional needs (like surgical training).

  4. #4 John C
    March 1, 2010

    Great post.

    The closest we are to biological simulation is the mapping of certain genomes and predicting genetic outcomes from the information. That’s a very, very, long way from live simulation of an organism.

    In fact, live simulation has some vast implications – an accurate simulation would involve using a subject that feels as much pain as it’s material analogue, only the medium it exists in would be different. Ethically, you might have just gone round in a circle. :)

  5. #5 JohnV
    March 1, 2010

    I think one of the issues at hand is that the people making these arguments are strikingly uneducated (and/or very stupid and/or deceitful). If they understood anything about computer simulations (they don’t) and they understood anything about the complexity of a biological system (obviously they don’t) they would realize just how laughable the suggestion that a computer could be used it.

    Gee lets take a simple bacteria system with 3000 proteins. How many potential protein-protein interactions is that? Homodimers or heterodimers? Maybe there are complexes (there are) of 3 or more proteins? What about interactions with metal ions and the like? Interactions with DNA and RNA? Other ligands? Membranes? Neighboring cells?

  6. #6 Eugene Day
    March 1, 2010

    I design and analyze Discrete Event Simulations for a living. I’m a D.Sc. Systems Engineer. I know a thing or two about simulation, and medical simulation. I have peer reviewed articles in press about them. The state of the art for medical/pharmaceutical simulation is called the Archimedes Model (I have no stake in it: it’s not mine and I do not specifically endorse it. I only point out that it’s the best I’ve seen so far.). It’s useful.

    The major problem with simulation is exactly what MarkCC says: they do what they’re told. Now: that doesn’t mean we can simulate complex systems which will hve surprising and useful results. Work done in Agent Based and discrete event modeling is getting more interesting all the time. But I wouldn’t trust for an INSTANT medical results based totally on simulation.

    There are surgical simulators that do a decent job, and show promising results. And some Ophthalmologic Simulators are promising to obviate the need for Sheep’s Eyeballs totally. But it is still being researched. And think of this: We need to use sheep’s eyeballs to determine if we no longer need sheep’s eyeballs.

  7. #7 Shawn Smith
    March 1, 2010

    And hasn’t your recent series on Chaos theory pointed out another problem with these models? If they are truly modelling a chaotic system, then how would anyone possibly say with any certainty that the simulation is “correct”? Also, I can’t remember where I read it, but I remember reading a headline that said researchers had finally completely modelled a proton or a hydrogen atom “properly.” The implication was that it wasn’t a simple task. If something so fundamental takes huge resources, then one can only imagine how complicated a model of a single cell would be, let alone an organism.

  8. #8 dete
    March 1, 2010

    JohnV: You’re right, but you’re actually directly opposing one of Mark’s points, in a sense!

    Mark says: Computation resources aren’t the problem, we don’t even know how these things work, so we can’t simulate them.

    You say: We know how some of these things work, but we don’t have the computation resources to simulate them.

    You’re both right! (Well, except Mark’s point about CPU horsepower not being a limiting factor. Look at Folding@Home. Where we do have very good models, and we are still well and truly running up against the limits of how much processing power humanity has available.)

  9. #9 James Thompson
    March 1, 2010

    @JohnV – You’re absolutely right, and I’m fairly certain that we can’t even model individual proteins very well without experimental information, either from a protein of interest or a homologue. To a very rough approximation, a protein backbone can be in one of 5 states. This means that for a 100 residue protein, we need to search 5^100 different conformations, which is far too many for any reasonable computation. This neglects the side-chain conformations and modeling those in will make the simulation even more computationally demanding. Even worse, most proteins that we’re interested in looking at are significantly larger than 100 residues.

    Your point about complexes and other interacting subunits is a really good one, and even at smaller levels of organization we run into huge barriers.

  10. #10 dete
    March 1, 2010

    One thing that just occurred to me: There’s an analogy between the argument that we can’t do these simulations and the ID argument that information can’t be created by mechanical systems.

    We can’t create simulations of living beings on computers because we (the creators) don’t have all the information needed to insert into the system. That sounds an awful lot like saying that unless a creator puts information into the system, the system can’t produce that information…

  11. #11 JohnV
    March 1, 2010

    In regards to an intelligent design implication of this, my response would be: I suppose given enough time and computational power (a billion years? a large planet full of “computers”) you could just have your computational system randomly assign values to all the variables and eventually an accurate model could be produced through an iterative process? :p

    I wasn’t especially trying to suggest that once fully recognized the problem becomes too big to simulate, simply that there are so many unknowns that we won’t be able to ever (in the near future) be able to full recognize it.

    I mean, you can experimentally approach the protein interaction question through use of a high throughput yeast-2-hybrid system (and people do that). The shortcomings of that are it only models interactions between 2 proteins on a 1:1 basis and like any system it generates false positives and false negatives.

    I’ll leave the estimations on computer power being sufficient to you guys since I have no idea.

  12. #12 eric
    March 1, 2010

    Dete I don’t see the connection. The problem here is that nature has put more information into the animal systems than we can currently understand.

    No one’s claiming the insides of test animals must be designed because we can’t understand them. That’s about as classic god-of-the-gaps as it gets.

  13. #13 DaveM
    March 1, 2010

    Great explanation – as I like to use as an analogy, “if it were so easy to create a predictive model for something as complex as a biological system, why then doesn’t Google maps yet know about the mini-market around the corner that opened two months ago?”

    Computer systems are great for analyzing data, but they cannot create data sets where none currently exist, and any data points that are drawn from a model by correlation require experimental validation in a functional biological system.

  14. #14 Coriolis
    March 1, 2010

    Being a physicist doing quantum atomic simulations, I wouldn’t say computing power isn’t the problem. If you could actually simulate biological systems at the level of fundamental physics, things could be different. Of course the current limit even with not very accurate methods is at the small protein level (and if you’re using the type of methods which are really accurate it’s a couple atoms at best).

  15. #15 Gray Gaffer
    March 1, 2010

    I don’t think we know enough about our biology to say for sure its a chaotic system, although I suspect it is so. But, for example, there are more possible interactions between neurons in a single brain than there are atoms in the observable universe, so that right there says (to me) we can never – repeat never – regardless of the state of our certainty about our knowledge, write a computer program that will correctly simulate a human being.

    But what we can do with computers is still pretty amazing. Regarding “For example, we can do wind tunnel simulations of aircraft designs”: nerds, check out X-plane. It’s a flight sim program that flies from the Physics of the aircraft model. You can design your own planes and see how they fly. It is so good there is even an FAA certified training version of it. And a Mars flight vehicle. SST. 777. etc. My buddy has three Mac Minis synched up with large monitors on each for a full cockpit view, enough to induce vertigo (three machines simply for rendering power within his budget). They practically give it away. (no connection, just wowed by it)

  16. #16 Anonymous
    March 1, 2010

    and this is a great lead-in to a discussion of climate modeling (and global warming, etc) (please!)

  17. #17 Mark C. Chu-Carroll
    March 1, 2010

    @7:

    I don’t think that chaos is a problem.

    An awful lot of how the basic processes of a cell work are pretty much chaotic by definition. For example, a key part of how ribosomes can “read” RNA in order to produce proteins involves brownian motion. But when you scale up a little bit, in the aggregate, things become relatively ordered.

    That aggregate order is fundamental to how medicines work. Effective medical treatments are pretty much by definition the things where you know what they’re going to do. Give a person with a bacterial infection a reasonable dose of antibiotics, and you know that it’s going to kill the bacteria but not the host cells. Given a persons weight, you can predict how much antibiotic you need to give them. Give a person with a back injury an anti-inflammatory, and you know that it will reduce their pain and inflammation.

    If you can’t predict how a body is going to react to a particular treatment, because the response to that treatment is chaotic, then it’s not going to be an effective treatment. Chaos means that you really can’t predict what something will do; if the response to a treatment is chaotic, that means that you don’t really have any idea of how the patient will respond. And that’s not good medicine.

  18. #18 hibob
    March 1, 2010

    @dete #8: I’d say folding@home suffers the same problems as almost all other simulations of proteins: the model isn’t good enough to yield useful predictions much of the time AND processing power is limited.
    We still suck at making de novo (made from scratch) models of proteins using what we know from physics and chemistry. If you use the best, most detailed mathematical models of how atoms interact and then try to scale up to the level of a protein in a box of water, you end up with a model that A, takes to long to calculate to be useful, and B, doesn’t resemble the real world.
    So the models include a lot of fudge factors, both to knock down the computational load and to deal with the inaccuracies of the de novo models. Even plain old water is hard to model accurately enough when it comes to interactions with proteins and DNA, so most water molecules are represented by bulk coefficients instead of actual atoms (Folding@home added Gromacs, but still … ). Using full non-linear Poisson-Boltzman equations to recalculate partial charges, etc, at every time step is a lot of calculating, so often the force field uses simplifications when and where (one hopes) the full model isn’t necessary.
    The result of all the fudge factors is that we end up with models that can’t really predict results that are much different from their starting states.

  19. #19 Sparky Clarkson
    March 1, 2010

    The simple fact is that we can’t even get particularly accurate binding predictions for proteins for which we have a great deal of structural and biochemical data. Nor can our simulations predict the conformation of a single protein over time periods as long as a microsecond without consuming mind-bogglingly vast quantities of processor time and storage space (or accepting drastic simplifications). And aside from a few examples, even those simulations have not been extensively validated. To propose that we can at present or in the near future somehow predict unknown molecular interactions for the billions of proteins and membranes in a whole organism for which we don’t know many of the protein conformations (even for those proteins that have deep conformational minima), or the expression levels under non-pathogenic conditions, or the protein-protein interactions (keep in mind that these would need to be QM simulations in order to allow for chemical reactions so we could trace the effects of drug metabolites), to say nothing of the details of each gene’s regulation, and that it would be possible to extend this simulation over the weeks of time that would be needed to assess the new equilibrium in the organism, is so incredibly absurd it’s not even laughable. Next the AR crowd will be proposing that we can ask the magic medicine pixies whether our new drugs are safe and effective.

  20. #20 jöbin
    March 1, 2010

    You guys are actually making this much more complicated than it all is. Assuming we have enough computing power, and we completely understand all of biology.

    Go back to your airfoil plane problem. As accurate as these tests are, no one will load a new passenger plane full of people and send it off from just the results of simulations. The idea is completely ludicrous. There will ALWAYS be human error, and the plane will need to be test flown.

    Simulations can give fantastic incite and possibly reduce real life testing, but they will always be necessary…

  21. #21 daedalus2u
    March 1, 2010

    Biological systems are chaotic. That is the ideal place for a biological system to self-regulate itself because it takes differential effort to go from one chaotic attractor to another.

    All natural neural networks self-regulate at the near percolation threshold. That is a true critical point, where the properties of the network change exponentially with connectivity. Again, that is the idea place to self-regulate at because it takes differential effort to get to another place. A single neuron firing can change the whole state of the brain. Can’t get much more non-linear than that.

    Of course it is controlled chaos. How exactly is it controlled? Very very carefully. With figurative flapping butterflies all regulated to flap (or not) at the right time, place, direction, magnitude, etc.

  22. #22 Daniel
    March 1, 2010

    I’ve been a reader of this blog for quite a while now, but I have to say that I am very disappointed with these recent posts. I work in Computational Biology and I believe this discussion is mixing two important points.

    1st point: Is animal testing important ?

    Yes, because as Mark said, garbage in garbage out. We need real data to build models. However, because of that same reason, the data required to build an accurate human model is human data. Of course, there are animals with genome and metabolism similar to ours, but the difference is still enough to make the simulations inaccurate for humans. A good human model should consider the sequenced human genome (already available), in vitro data of human biochemical kinetics, and validation with real human cases.

    2nd point: is animal testing right?

    From my perspective, no! The fact that something might be useful, doesn’t make it right. And (non-human) animal testing is wrong, by the same reason that involuntary human testing is not allowed. Because no one should be forced to something against their will, independent of the species that one belongs to, homo sapiens or any other.

    Mark, I’m glad science was able to find a cure for your problem. But I have to argue that your life is as important as the life of the animals that where used in those tests, and as of any other sensitive creature.

  23. #23 dete
    March 1, 2010

    Eric@12: I think you’re looking at it from the wrong angle. Dembski’s (and friends’) argument against evolution is that mechanical systems can’t produce information; any information produced by a mechanical system must have been seeded somehow by the originator/creator of that system. (I don’t buy this, BTW, but that’s the argument.) (BTW – This is all predicated on the notion that biological systems are essentially very complicated mechanical systems, that’s a pretty key point!)

    I’m just pointing out an irony: I know MarkCC doesn’t like Dembski’s argument any more than I do, but to an untrained (or unsubtle) mind, his statements are very similar: We can’t get the information we want out of our simulators because we don’t have the right information to put in. MarkCC is almost (almost!) saying the same thing as Dembski here: The creator has to put the information in in order for it to come out.

    The difference, of course, is that Dembski claims that the information that comes out of a mechanical process must be less than or equal to the information that went in; the mechanical process can regurgitate information but not create it. With the simulations of the sort Mark talks about, the information coming out depends on, but can be very different from, the information that goes in.

    In terms of evolution, the information going in is randomness (which doesn’t intuitively feel like information at all) and the current state of an organism’s environment. The information coming out (an organism that can succeed in that environment) depends on that input, but is very different from it.

  24. #24 sharky
    March 1, 2010

    Daniel @ 22: Okay, let’s momentarily grant this point…

    Which would you pick?

    I’d also like to add that even if what happens to one cell could be simulated, then what happens to that one cell can only be calculated as one scenario. How cell groups and tissues work together is far more complex and mysterious than even one cell alone, and trying to predict one cell alone, without the rest of the organism, could be like trying to play dominoes by knocking the start of the chain over the wrong way when it comes to actually getting results.

    I try to buy cruelty-free products by even avoiding products with a chemical that’s been tested on animals (apparently some companies avoid the “not tested on animals” label by not testing *that* product, but its ingredients.) When it comes to medical experimentation, I can only support doing what we have to as responsibly as we can.

  25. #25 Brian
    March 1, 2010

    On two occasions I have been asked, “Pray, Mr. Babbage, if you
    put into the machine wrong figures, will the right answers come
    out?” In one case a member of the Upper, and in the other a
    member of the Lower, House put this question. I am not able
    rightly to apprehend the kind of confusion of ideas that could
    provoke such a question.

    [from "Passages from the Life of a Philosopher", by Charles Babbage]

  26. #26 hibob
    March 1, 2010

    @#22Daniel:
    “However, because of that same reason, the data required to build an accurate human model is human data. Of course, there are animals with genome and metabolism similar to ours, but the difference is still enough to make the simulations inaccurate for humans. A good human model should consider the sequenced human genome (already available), in vitro data of human biochemical kinetics, and validation with real human cases.”

    Yet somehow a panel of tests in several animal species is still more accurate than our best computational models of humans when it comes to testing new chemical entities. If we already have human biochemical kinetics and ADME data for the molecule necessary to make a good human model, the model would just be telling us what we already know.

    Please, cite me a human computer model that is more accurate than a panel of animal studies in predicting the activity of a new chemical entity.

  27. #27 Doug Spoonwood
    March 1, 2010

    I have to agree with the crowd that says that computational power does come as serious difficulty with relying on such simulations (in addition to Mark’s point). Even the simplest biological systems come as massively complex. From simple considerations, if we change just an electron or two in a biological system, we change an atom, which in turn can alter molecular structure or whether that structure forms, etc. Bremmerman’s limit (from our current understanding of computation, based on quantum theory) http://en.wikipedia.org/wiki/Bremermann's_limit, fundamentally places limitations on computational power. Too many biological system problems of interest here exceed that limit. So, not only do we have the GIGO difficulty, but biological problems end up transcomputational, and thus in order to get accurate enough information, we need real-world animal tests. That such comes as ethical or even useful enough to live with ethical difficulties which such testing produces, comes as a different story.

  28. #28 Flaky
    March 2, 2010

    @Daniel(#22): I’m glad you brought up the real argument behind the obfuscating BS that animal rights activists usually put forward. If you remove the nonsense surrounding the issue, at the bottom line really lies the question if it’s ok to hurt and kill animals in order to improve or save human lives. Considering that, it becomes difficult for me to understand why there is so much focus on animal experimentation by the ARA crowd, when people routinely kill animals for food, often simply because people enjoy eating meat, rather than out of necessity.

    I suppose it’s harder to sell that people should stop eating meat, when the way nature works is that animals kill other animals for food. (Not to mention that while people can survive on vegan food alone, there are good reasons why they shouldn’t.)

  29. #29 Anonymous
    March 2, 2010

    Not only that, but in at least some cases the animals we eat for food are thriving exactly because we want to keep them around to eat. I’d be willing to bet that the average lifespan and general health of domestic cattle is significantly better than their wild kin who are scrounging and being chased down by wolves. Also, we worry about how to kill them humanely, while the wolves literally eat them alive (to begin with anyway).

  30. #30 Necandum
    March 2, 2010

    With the ethics of animal testing, the fact remains that a human life is more valuable than that of an animal’s. In any sense you care to look at it, whether that be economic, social or intuitive. Note, however, that I am not talking about pain. I think there is no greater evil upon this earth than cruelty, but death is an entirely different beast.

    So, if during experimentation, animals were to be treated with utmost care so as to avoid causing any pain or suffering, what objection could then be put forth?

  31. #31 metus
    March 2, 2010

    I would like to know how those guys who would like to build a biological system up from atom/molecular level (given enough computing power) want to identify the initial conditions of those billions and billions of molecules.

    @Daniel
    You have to take into account the degree by which models fail reality. No model is perfect.

    Animals chosen as models for humans are often correct enough to produce good results. The error is tolerable.
    On the other hand, todays biomathematical models on cell level are often as bad as random guesses. Overparameterized, inaccurately measured and based on unrealistic assumptions.

    Considering morals:
    People like me will give people like you the right to avoid animal experiments.

  32. #32 James Sweet
    March 2, 2010

    Suppose that a simulation of a human body, sufficiently detailed to replace animal testing, requires simulating all physical interactions at the atomic level…in that case there just isn’t enough computing power in the world to do that kind of simulations. And there isn’t going to be in quite a while.

    Not “quite a while”, more like “ever”. That’s a whole lot of freakin’ atoms. It’s dubious that such a thing could ever practically be simulated.

    Anyway, you start to get into philosophical/ethical issues simulating at that level. If the simulation is capable of producing the phenomenon we recognize as “consciousness”, wouldn’t experimenting on such a simulation have the same ethical problems as simulating on a human?

    There’s an interesting discussion of the subject in the comments to this post.

    ANYWAY, these are all nits that don’t undermine the basic point of Mark’s post. I think when Mark said computational power wasn’t a problem, he meant that there is far more computational power available than we are able to exploit when it comes to biological simulations. I’m not sure I agree that that is always the case, but the basic point is valid.

  33. #33 Pat Cahalan
    March 2, 2010

    > Even the simplest biological systems come as
    > massively complex. From simple considerations,
    > if we change just an electron or two in a
    > biological system, we change an atom, which
    > in turn can alter molecular structure or
    > whether that structure forms, etc

    This doesn’t necessarily apply. In fact, it likely doesn’t apply. In fact, it almost certainly only applies some of the time :)

    Complexity doesn’t necessarily mean what most people think it means. Most complex systems (including all biological systems) have a fairly substantive ability to correct themselves at a higher level of abstraction than what you’re talking about here.

    For example, sunlight hits your skin. Some of the UV radiation damages a chromosome in one of your skin cells. You can get skin cancer from this. However, the skin cell naturally dies and is shed. Or the skin cell turns cancerous but is destroyed by your body’s immune system. Or the skin cell’s chromosome is altered, but that results in the skin cell’s inability to sustain its protein wall and it dies.

    Computer models are verifiable only *probabilistically*, absolutely. But it’s also true that biological verification via experimentation is also only *probabilistically* correct.

    A useful computer model of a complex system is useful when it can be used to predict behavior better than a random choice. A very useful computer model of a complex system is one that can predict behavior of that complex system better than some other method.

    It will *still* not be a perfect model, any more than any lab animal experiment is going to yield “truth” about the safety of a medication. The “animal testing only approximates human conditions” is a poor point to make against animal testing -> ALL methods of scientific inquiry only approximate the conditions that they’re attempting to expose to observation.

    The question is, how well does the method of inquiry predict the outcome of a real world scenario?

    It’s never going to be a certainty that objective truth has been reached. If you want “proof”, “certainty”, or “truth”, stick to metamathematics and philosophy.

  34. #34 Phyllograptus
    March 2, 2010

    Very good post. As a geologist who has been involved with setting up the input data and models for fluid flow simulations I’ve found the post & descussions very interesting. Coming from a different background I have a question for the Biological & Medical simulators out there. One of the processes we often use is Geostatistics (Simplistically explained) where we create a model from the data we have (which tends to have a lot of variability and a large range of input values) and then use geostatistical software to probabilistically fill in the data we are missing. However this tends to create inputs that are extremely large (many grid cells with many input parameters per grid cell but with smaller ranges and variabilities) and overly complicated so we go back and upscale the geostatistical model to reduce the input and sort of give an “averaged” model to input to see if that “average” type model can explain the physical behaviour we observe. Do other numerical modelling simulators use similar techniques?

  35. #35 Anon
    March 2, 2010

    This whole post is a rather lengthy straw man argument. You end your essay by making the claim:

    “So simulation absolutely has a role. And hopefully, it can reduce the number of animals that are used in medical experimentation. But it can never replace them. There really is no substitute for reality.”

    There is no major (perhaps even minor, I don’t know) animal rights group that makes the claim that computer simulations can functionally replace animal testing. Not even the poster you quote at the beginning of this essay makes anything more than the claim that “informing people about the alternatives to animal testing [such as computer simulation]” is important. No claim is made that computer simulation functionally replaces every aspect of animal testing, only that in some cases it can perform the same function as an alternative. (Of course, I cannot divine what the post meant; if by ‘alternative’ s/he meant complete replacement then his/her view would be rather marginal even within the animal rights movement.)

    As for animal testing itself, surely one has to admit it is a complicated issue depending perhaps more on moral impulse than scientific fact. Certainly the relevant scientific have to be interpreted morally in order to make a moral decision about the vivesection. (Since you seem to be in the habit of misinterpreting statements, by this I do not mean that scientific facts are irrelevant, only that they by themselves say nothing about right and wrong.) If you attribute the same rights to animals as you do humans, you have to ask yourself whether it would be justified to torture one human against his or her will in order to possibly cure a disease for thousands of people. Primates are still used for experimentation, and there are scant safeguards in place to ensure as little pain as possible for the animals — or even transparency about what is being done to them.

    None of this, of course, is to defend the animal rights vigilantes who have taken to threatening someone’s children. Indeed, such actions are perhaps a major reason for a lack of transparency in animal experimentation.

  36. #36 Jon H
    March 2, 2010

    ” Considering that, it becomes difficult for me to understand why there is so much focus on animal experimentation by the ARA crowd, when people routinely kill animals for food, often simply because people enjoy eating meat, rather than out of necessity.”

    Even worse is people in America or Europe, who are from Africa, who smuggle in ‘bush meat’: the meat of jungle wildlife such as apes and monkeys, or other possibly endangered species. It’s purely out of nostalgia, not a matter of survival.

  37. #37 anon
    March 2, 2010

    A friend pointed out to me that while PETA does not seem to explicitly make the claim that animal testing can be replaced entirely by computer simulation, they do in at least one place obscure this point.

    http://www.peta.org/ABOUT/faq-viv.asp

    All the same, this obfuscation by PETA aside, the view that computer modelling could replace animal testing in every aspect — if it is held by anyone — must be extremely marginal. Animal rights activists of course advocate computer simulation when applicable, but few have the illusion that vivisection is redundant and done because processing power is more expensive than chimps.

  38. #38 metus
    March 3, 2010

    @Anon

    Computer simulations are cheep. If they become correct enough, they will be used. Computer simulations are used in many different areas in engineering already without pressure groups who push them.

    Our current problem as I see it is that computer simulations are pushed into areas where they are known to fail. IMO it is immoral to experiment with animals to produce models for computer simulations that you know are going to fail.

    Animal rights groups are not helping in this regard. They should focus on improving conditions under with lab rats are kept, instead of tracing fantasy solutions.

  39. #39 metus
    March 3, 2010

    @Phyllograptus

    On cell level, biologist try not to use spatial information, because there is no reliable way to measure them. So concentrations are averaged along space. You usually can’t measure many important molecules in living cells at all, so they are estimated (least square regression) according to a parameterized model and observations of other components. This is in bacteria. (Boy, we can’t even handle bacteria! How should we produce models for humans?)

    In animals you usually try to look at many dead cells instead. Which is completely ludicrous as well. You have to compare molecules (averaged over many cells) from one animal for time step A and use those of a different individual for time step B and so on …
    Do I need to mention that measures of these quantities are not very precise and that biologists have a hard time to produce just qualitative information?
    Those molecules are inside of cells inside of tissues. They have to be extracted and extracted and separated from other molecules before you can even start measuring them. Not every measurement will be “successful”, so you will end up with missing values in your grid. So you can only estimate them.

    OK, when you have got your data, you are usually confronted with enormously large complex interaction networks ― that are only proposed (!) models (based on insufficient observations and insight). When there is interaction with DNA you may get systematic stochastic effects. And yes, I have seen scientist actually average them as well.

    In the end, you need enough parameters and speculative models to come up with something that fits your observations or some one who fits his observations to your model. If somebody else really tries to reproduce your results (which is expensive and won’t result in publications), he will work under different conditions with, different individuals …
    Sometimes you might be lucky that your system is not complex, that cells are easy to handle and so on.

    Physiological and anatomical systems are large enough that you can produce reasonable data.

    Neuro-science is a mixture. Apart from some waves, I don’t expect much here. (Marc, I think you will have to hope on good old fashioned medical reasoning. I doubt that any computer simulation will come up with an answer that will cure your depression.)

    Gene expression: I haven’t seen anything out there that I would even consider a serious attempt of a model. It’s all machine learning without a clue.

    Evolution: Even though I trust those old bones from field work more than any theoretic model, you can have a look a different scenarios to get a feeling of what is at work here.

  40. #40 mike
    March 3, 2010

    From 22: “But I have to argue that your life is as important as the life of the animals that where used in those tests, and as of any other sensitive creature.”

    From 30: “With the ethics of animal testing, the fact remains that a human life is more valuable than that of an animal’s. In any sense you care to look at it, whether that be economic, social or intuitive.”

    I think this difference of opinion is driving the animal rights debate. I sympathize with the human exceptionalism side but have trouble logically supporting it. Why is it so clear that an animal’s life is not as ‘valuable’ as human life?

  41. #41 Phyllograptus
    March 3, 2010

    @metus
    thanks for the response. It looks like the biological & medical modelling community have very similar problems and use fairly similar methods to deal with them, i.e stochastics etc. Interesting to learn about the similatities across diciplines.
    One slightly nasty Schadenfreude I have is watching a newby modeller present you their model results with that glowing enthusiasm and absolute faith they have in the results and listen as they say that they have “THE ANSWER” and then watch their faith in “THE ANSWER” crumble as you point out the problems with the data, the assumptions (sometimes even the theory) and then you make them go back and run another simulation using slightly different but still valid data ranges and get a different “ANSWER”. Then you get to the pleasant part of the learning process and watch as they learn about the applicability and limitiations of numerical models in the scientific process.

  42. #42 DrFrank
    March 3, 2010

    I’d have to agree with Flaky at #1, computational power still *is* a problem, as an atomic simulation would be pretty much a necessity to actually identify unpredicted chemical reactions, etc.

    Horrifically simplistic lower bound estimate follows:

    * Approximately 10^28 atoms in the human body.
    * 10^10 computers available (optimistic assumption of more than one per person on Earth)
    * This means each computer would be simulating around 10^18 atoms.
    * Assuming an entire atom’s state change can be evaluated to a sufficient accuracy in one clock tick (hahahaha) with a 1 THz processor leaves 10^6 seconds per simulation step
    * This would mean that one simulation step would take about 11 days.

    And, of course, the simulation step would probably have to be around the order of a picosecond in length…

    So yes, I would say there’s a good 12 orders of magnitude increase in computing capability required before we could say that complexity wasn’t a problem, and probably a lot more in reality.

  43. #43 Pat Cahalan
    March 3, 2010

    @ DrFrank

    > I’d have to agree with Flaky at #1, computational
    > power still *is* a problem, as an atomic
    > simulation would be pretty much a necessity to
    > actually identify unpredicted chemical reactions,
    > etc.

    An atomic simulation would most likely not be necessary or advisable.

    In order to model something properly, your level of abstraction must match that which you’re trying to model. Pushing the model into a lower level of abstraction is unlikely to increase the model’s predictive accuracy much, and is very likely to change your boundary conditions to the point where your model is unmanageable.

    Again, you are not looking for truth, here. You don’t need to know that in any possible set of circumstances, given any possible patient and any possible intervention, that the probability of adverse reactions is zero. That’s a completely non-feasible target. You can’t get that with *any* method of inquiry, even experimentation.

  44. #44 DrFrank
    March 3, 2010

    The basic question that such a model would need to answer is “will compound X have any likely adverse reactions?” – what sort of level of abstraction would you recommend to deal with this in order to retain any meaningful generalisability?

  45. #45 G.Shelley
    March 3, 2010

    Of course for a quick refutal, one can just ask the animal activists for a solution to the protein folding problem, since that must by necessity be solved for computer simulations to replace drug testing on animals.

    Or ask them to use a computer model to find the function of every protein identified from ORFs in the human genome and which proteins they interact with. Or even just under what circumstances it is transcribed and what if any RNAi is involved.
    On another point, the original post would only be a straw man if people didn’t claim that computer modeling can do all that animal testing can do.

  46. #46 Pat Cahalan
    March 3, 2010

    @ DrFrank

    > The basic question that such a model would
    > need to answer is “will compound X have any
    > likely adverse reactions?”

    That’s not a basic question.

    Define “likely” and “adverse”. 1 patient in 10,000,000 possible death or disfigurement? What are acceptable “adverse” reactions? Is “headache” an acceptable adverse reaction? To heart medicine, maybe. To a headache remedy, probably not.

    As Mark pointed out, in order to validate your model, you need to do some sort of confirmation. You do background investigations into likely causes for an ailment. You design an intervention which possibly will address the cause. Depending upon your understanding of the ailment and the mechanisms, you probably need to validate your model design with additional investigations, most likely experiments. I’m not arguing that you don’t need experiments.

    I’m saying that your computer model that pops out the other end can possibly be useful in other cases (again assuming your understanding of the mechanism is correct). Computer modeling is a complementary investigation method, not a replacement. But experiment is itself also a complementary investigation, we just didn’t have very many things to complement it with before. We’ve only ever been able to give probabilistic predictions as to outcomes outside the experimental population. Experiments don’t automatically generalize either :)

    Optimally, for really strong results in any scientific endeavor, you want multiple confirmations. You want a foundational theory that is causal. You want experimental validation. You want computer models of the behavior. You want epidemiological verification across large populations. You get all those things, and you’ve got a pretty well understood phenomena. You don’t have all of those things, you might still have a useful intervention, but the mechanism isn’t understood, and you still have some grounds for inquiry.

  47. #47 johnx
    March 4, 2010

    Humans are the best test model. Test on humans. That way when the space aliens land and want to test on us, we can tell them to test on themselves like we do and they will totally appreciate that argument, like we do.

  48. #48 Robert
    March 4, 2010

    The one situation where computer simulation could be extremely useful, and well within our capabilities would be to answer questions such as:

    “Will this artificial heart valve properly open and close in a blood stream”

    Such a simulation could identify a lot of problems before actually implanting the valve (or any other physical implant)

    A second step might be to make a physical model of the heart and compare these results with your simulation.

    Finally, since computer or physical models are never perfect, you’d need to try the valve in a real heart. Furthermore, this kind of model will never tell you whether immune responses, scar forming, etc… will pose problems, this is another reason to always test on a complete animal before implanting in humans.

  49. #49 folbec
    March 4, 2010

    Yet another crackpot honeypot post.
    You only need to add Cantor’s diagonalization to perfect the trap.

    I love this blog.

  50. #50 Sparky Clarkson
    March 4, 2010

    An atomic simulation would most likely not be necessary or advisable.

    I disagree. Not only would an atomic simulation be necessary (on at least some scales), but quantum mechanics (as opposed to computationally less-demanding molecular mechanics) simulation of at least some molecules would be required in order to identify unexpected reactivities. For any given drug, of course, the vast majority of the system can be coarse-grained, perhaps even to the same degree as a typical molecular biology slide (proteins as Euclidean solids), but we can only know which portions of the system can be rendered schematically after we know what the drug and its metabolites interact with in vivo. This knowledge, obviously, is what the simulation is intended to acquire, leaving us in a bit of a catch-22.

    Animal experimentation allows us to determine drug safety phenomenologically – we don’t have to know how effects occur, we simply have to observe that they do. Simulation requires us to investigate drug effects mechanistically – we can only generate accurate predictions if we correctly assess the entire chain of molecular interactions. Without some a priori knowledge of proteins the drug/metabolites absolutely will not bind to under physiological conditions, coarse-graining cannot be employed. Even then, there will be the challenge of correctly integrating the coarse-grained portions of the simulation into the atomistic portions, given that the energies are guaranteed to be different and may not be linearly (or even consistently) related.

    Also – don’t forget that we would have to simulate the water and small molecules (ions, sugars, nucleotides, cofactors, etc.). In a system this large that challenge is not inconsiderable, but simulations in implicit solvent just aren’t to be trusted, and the influence of ions on binding events in particular cannot be disregarded.

  51. #51 Jon H
    March 4, 2010

    “The one situation where computer simulation could be extremely useful, and well within our capabilities would be to answer questions such as:

    “Will this artificial heart valve properly open and close in a blood stream”"

    Actually, the even simpler situation where simulation is useful would be those mostly concerned with forces, for example, in calculating the forces acting on an artificial limb or joint during various activities.

    That’s about the simplest case of human body simulation I can think of, apart from digital crash test dummies.

  52. #52 Daniel
    March 5, 2010

    Just a reply to the replies on my comment…

    I do not believe that at the moment, computer simulation is a replacement for animal testing. With some effort, it may be in the future. However, to build a good human model, the best animal for testing is the homo sapiens.

    And why don’t we use homo sapiens for testing? Because neither you or me wants to be tested upon. Why is it ok to discriminate and test on other species then?

    Some argue that we already eat animals and it’s basically the same thing. That’s true, animals eat other animals, its how nature works. But the way we industrialize the whole process adds unnecessary suffering to it. Not claiming that we should all go vegan, but reducing consumption would be a good thing.

    If you read about some of the bizarre experiments that have been performed on animals, it is a cruelty far beyond anything that occurs in nature.

    #40
    “I sympathize with the human exceptionalism side but have trouble logically supporting it. Why is it so clear that an animal’s life is not as ‘valuable’ as human life?”

    I believe that’s by the same reason your family’s life is more important to you than a stranger’s life. It’s just emotional, however it still doesn’t make it right.

    In summary to the objective of this post. I agree with Mark that what those activists did is just wrong. It’s even wrong to call them animal rights supporters.

    The animal rights concept is based on the non-discrimination among species. Therefore an animal rights supporter cares for homo sapiens as well as for other species, and would not try cause arm to people.

  53. #53 ponderingfool
    March 5, 2010

    We still suck at making de novo (made from scratch) models of proteins using what we know from physics and chemistry. If you use the best, most detailed mathematical models of how atoms interact and then try to scale up to the level of a protein in a box of water, you end up with a model that A, takes to long to calculate to be useful, and B, doesn’t resemble the real world.
    ********************************
    Of course if you even understand folding in a box of water, you are far off from what is happening in the cell. The cell is dense (E. coli: 300-400 g/L of macromolecules), so crowding effects have to be taken into account which favor the folded state as well as misfolded states & aggregates.

    As for the idea we should try and not harm animals taken to its logical conclusion would mean that humans should cease as a species. We know we kill animals all the time . I Even the strictist vegetarian ends up killing animals. Look at the estimates of what we kill while we sleep over a lifetime. If true equality, then how can we let our species continue? We know a slaughter is happening but we let it continue. If we truly thought humans were equal to animals (and why stop at animals?, what of the rights of other living organims, why stop at life that is most similar to us humans?), we would stop reproducing and let ourselves go extinct.

  54. #54 Daniel
    March 5, 2010

    #53

    A threshold, great point! That’s a very good and valid argument you can pose against animal rights supporters. Why stop at dolphins, dogs, monkeys and rats? Lets protect the rights of insects as well, and maybe every living thing?

    You can use the argument that we eventually need to kill other species to survive. That’s true, but lets put the question the other way around. Should we kill some among our species in order to survive as well ? Why set the threshold between our species and the next one? Lets use homo sapiens for experiments, and spare a hundred lives to save a thousand. Also seems logical to the survival of the whole group.

    If we claim to be superior to other species, the minimum we can do to prove it, is to put our good sense to use and don’t just do things because we can. The homo sapiens survived a long time without science and technology, and trying to improve it at all means is not necessary for our survival. In fact, in the recent years it seems that our unstoppable greed is the biggest threat to it!

  55. #55 metus
    March 5, 2010

    @Daniel,

    “With some effort, it may be in the future.”

    No.

    “And why don’t we use homo sapiens for testing?”

    We do.

    “But the way we industrialize the whole process adds unnecessary suffering to it.”

    Ever seen tribal people kill an rhinoceros?

    “If you read about some of the bizarre experiments that have been performed on animals, it is a cruelty far beyond anything that occurs in nature.”

    Give me a break. What do you know of nature? You sound like a city boy. If you meet a bear, tell him hi from me.

    “Should we kill some among our species in order to survive as well ?”

    So you have seen that policemen wear guns?

    “Lets use homo sapiens for experiments, and spare a hundred lives to save a thousand.”

    I have a better idea. Let’s start with animals and use humans later on when we know more. Hey, that is what we already do.

    “If we claim to be superior to other species, the minimum we can do to prove it, is to put our good sense to use and don’t just do things because we can.”

    Boy, I would like to see you write a request for a an animal experiment: «Hey guys, I would like to use some animals just because I can …»
    Good luck!

    “The homo sapiens survived a long time without science and technology, and trying to improve it at all means is not necessary for our survival.”

    When does science and technology start? As soon as metal and a lab coat is involved? Somebody should have told the fire-guy thousands of years ago that he is just a dumb caveman and science would only start when the animal rights people feel like it.

    As I come to think of it, I have my doubts that science and technology is unnecessary for our survival, but I’m pretty sure that animal rights groups are. Nevertheless, they have at least the potential of improving our lives if they stop silly relativism.

  56. #56 Jonathan Vos Post
    March 6, 2010

    My PhD dissertation research was 1975-1977. I showed an extremely sophisticated mathematical way to speed up the high-resolution simulation of a single living cell by a factor over over 1,000. I’ve continued to publish on that subject since then. Now, a third of a century later, we are getting close to being able to fully simulate one living cell. Several teams in several countries are explicitly in the race. How long do you think it will take before we can do a high-resolution (space and time) simulation of a higher organism? Until then, in vivo and in vitro experiments are the only way to proceed, as in silico remains over the horizon. By the way, this also dashes the short-term hopes of “rapture of the nerds” followers of the Cult of Singularity who expect to “upload” their brains into software.

  57. #57 Pat Cahalan
    March 7, 2010

    > Animal experimentation allows us to determine
    > drug safety phenomenologically – we don’t have
    > to know how effects occur, we simply have to
    > observe that they do.

    No, it doesn’t. Animal experimentation allows us to approximately identify likely drug safety, which is why you have human trials. Even human trials only allow us to approximately identify likely drug safety, which is why you have drug recalls. Neither necessarily tells you why the damn thing is working, which Mark points out above.

    Even probable drug safety, once established, can be re-evaluated based upon risk, which is why you don’t take the smallpox vaccine any more. The more accuracy you have, the less ability to generalize you have. This is true *regardless of your method of inquiry*.

    > Simulation requires us to investigate drug effects
    > mechanistically – we can only generate accurate
    > predictions if we correctly assess the entire chain
    > of molecular interactions.

    No, we get more accurate predictions the closer we asses interactions. We are never going to have completely accurate predictions, because we can never simulate every person’s blood chemistry. The number of independent variables is too large. Our ability to properly identify the independent variables is limited. However, not all of the independent variables are going to matter under practical conditions, most of the time.

    There are lots of places where you can use simulation effectively without understanding the entire chain of molecular interactions in use. You can find out that mixing drug A with drug B in most patients is probably a very bad idea. You can find out that, given that your understanding of the mechanism is correct, you will see approximately one level of efficacy. Then you can test your assumptions experimentally and find out if your model approximates experimental results.

    But you’re not comparing apples to oranges. Experiments do not give you absolute truth. Epidemiological studies don’t give you absolute truth. Computer models don’t give you absolute truth. None of these tools needs to be only used under perfect conditions, because there are no perfect conditions. You’re not going to get deterministically infallible knowledge out of any single sort of scientific investigation. The best you’re going to get is a higher degree of probability.

  58. #58 metus
    March 7, 2010

    @ Jonathan Vos Post

    “Now, a third of a century later, we are getting close to being able to fully simulate one living cell.”

    Give me a break!
    Which cell shall it be?

    It is completely irrelevant how fast your computation will be.
    Garbage in garbage out.

  59. #59 Daniel
    March 8, 2010

    I think the discussion is focusing too much on simulation. Whole-cell simulation is not fundamental to perform biological studies. There’s a lot that can be learned by analyzing network reconstructions built from genome annotation. Moreover, everyone has a different metabolism, that’s why people have different diseases. Running a simulation yields a specific result that cannot be generalized.

    @Metus
    ” Boy, I would like to see you write a request for a an animal experiment: «Hey guys, I would like to use some animals just because I can …» ”

    I don’t know where you live, but I live in a place where bull fights are considered national culture. And you don’t need a lot of trouble to kill an animal around here. I know undergrads that have access to animals, and kill a bunch of them in the lab just for fun.

  60. #60 metus
    March 8, 2010

    Daniel,

    that sounds like a situation were you guys realistically can change something for the better.

  61. #61 T. Eugene Day
    March 8, 2010

    @58,

    I think that was Dr. Vos Post’s point: we are close to beinag able to fully simulate a system consisting of one cell. Now, if the simulation is properly done, it could probably be fairly malleable, and simulate any of a basic class of cell. But yes, the difference between a human neuron and a fungal spore is going to be so great as to make a single simulation attempting be able to make meanigful conclusions about both to be ridiculous. I suspect. I don’t know much about cellular simulation.

    Anyway, I think that Dr. Vos Post was pointing out just how limited the simulation is, not how powerful.

  62. #62 Sparky Clarkson
    March 8, 2010

    @ Pat

    You don’t seem to have gotten the point I was making about the two kinds of experiment, which was not that animal experiments give us certainty while simulations do not. The problem that animal testing “solves” (within whatever limitations you want to make about probabilism) is that it notifies us of ill effects arising from promiscuous molecular interactions, even if we have no prior knowledge (or no knowledge, period) of the interactions that give rise to adverse affects. Simulations can only “solve” the same problem if they correctly model the molecular interactions themselves at every phase from drug binding to gene activation. They must therefore be atomistic on at least that first level, or else useless. I’m not holding the position that animal tests give us complete certainty while simulations don’t. What I’m saying is that animal tests give us some confidence in this regard, and coarse-grained simulations will give us none.

    You claim that simulations could tell us about bad interactions between drug A and drug B, but how would we control for drug interactions that lie along some unexpected route? They might bind or react with unsuspected partners or activate compensatory mechanisms we don’t understand. Or they might even interact with regulatory mechanisms we don’t even know exist yet: there are a surprisingly large number of human proteins for which the catalytic activity isn’t even known. This is why so many drug interactions are detected only after a drug is approved for human use — the biology often works in unexpected ways. In an animal test the outcomes would at least be visible even if we have no idea of the mechanism. But a simulation would only work for us if we put in the mechanistic data to make it work.

    I am further doubtful of a coarse-graining approach because so many of the inputs would be garbage to begin with, even for proteins we “understand”. Of course, if we have accurate data about binding constants and turnover rates it should in principle be possible to use low-resolution simulations to model metabolite flow, transcription, etc. in a cell. But the data we have on these matters come from in vitro assays that were performed with the goal of getting data, period, not necessarily biologically relevant data. Our knowledge about protein binding, folding, and catalysis mostly comes from experiments performed at non-biological temperature or pH, in unusually clean solutions that minimize undesired cross-reactivity, with unrealistic salt concentrations, using substrates that may or may not be good models of the actual biological materials involved. The regulatory cycles proposed typically get assembled from data on single proteins or two-protein interactions, almost never a whole system running together in a tube. To propose that we can predict with any degree of confidence how “a cell” (a nerve cell? a cancer cell? a lung cell? concentrations of enzymes, metabolites, organelles, and regulatory proteins vary wildly) will react to even one novel toxin, let alone two, strikes me as extremely optimistic, almost to the point of hubris.

  63. Very nice post. That sounds that you guys can change something for the better.

    If we claim to be superior to other species, the minimum we can do to prove it, is to put our good sense to use and don’t just do things because we can.

    Great that you can pose against animal rights supporters.

    Thanks.

  64. #64 Jake
    April 12, 2010

    My good friend Sinead gives a really smart response to this nonsense: there is no scientific argument against animal testing, only an ethical one.
    http://asksciencedude.info/2009/10/15/podcast-episode-5-animal-testing/

  65. #65 Chris
    April 13, 2010

    I’m quite confident that I can articulate quite succinctly why we can’t produce sufficiently sophisticated simulations of the human body: because that would mean we could make humanoid robots.

    It’s an utterly absurd claim. Millions of different proteins, billions of different chemical reactions and we have fully understood…absolutely none of it.

    There are over 3,000 known genes involved in asthma! The ones that we do understand, we only have a crude, high-level understanding of — no mechanisms, just a basic idea of some of the cytokines involved. Which leads to me to the myriad of cytokines, which we barely understand, let alone have the intimate knowledge required to accurately model in an interactive system. I’m a former physicist turned medical student, I could be at this for hours.

  66. #66 Bert
    May 26, 2010

    Informative post!

    I firmly believe that there are surgical simulators that do a decent job, and show promising results. As Mark refers, in order to validate your model, you need to do some sort of confirmation. I agree that one needs to design an intervention which possibly will address the cause. However I’m not saying that you don’t need experiments. As far as simulation is concerned validation is most important thing to keep your simulation on track.

    The Simulation Medical is now providing training kits for the medical students which are designed by Doctors, nurses and educators. Its accessible to every student, to practice where they want, when they want, and as much as they need until they feel confident. It is both highly realistic and affordable. It Includes award winning interactive tutorial software – it’s like having your own private clinical tutor with you every step of the way.