1. Rick Wobbe on March 6, 2012 12:36 PM writes...

Isn't most biotech investment predicated on a strong belief in determinism (e.g. Genetic determinism, structure-function determinism, "past performance [of people] is a predictor of future success")? If it's misguided, isn't that kind of a discouragement to invest in biotech?

2. Nekekami on March 6, 2012 1:23 PM writes...

Investment is often misguided by the investors on their own, either by simply not fully understanding the subject or because they project their own wishes onto it, thinking that it's just a case of "you just need to work harder/throw more hardware at it".

As for the disciplines themselves, there are other strongly non-deterministic fields with emergent phenomena, such as serious psychology, which is lumped together with bunk psychology by the more ignorant physicists(who also tend to see anything that is not physics as "not rigorous/pure enough"), completely ignoring the fact that much of the math and research approach carries a strong resemblance to particle and quantum physics.

Ah, the fun of interviewing people when they respond different depending on the environment in etc... Or, from a friends studies, when two in a control group for a PTSD study turns out to be claustrophobic, necessitating a shift to larger rooms or even outdoor interviews, whereupon three turn out to be agoraphobic, and the list goes on... His comment was "I should switch to something easier, like solving Fermat's theorem"(this was before Fermat's theorem was solved)

3. PTM on March 6, 2012 1:38 PM writes...

"Isn't most biotech investment predicated on a strong belief in determinism..."

There can be no doubt that determinism reigns supreme in biology. Just look at how closely offspring resemble their parents - from cells to organs to whole body and their development stages - very little is left to chance.

The problem is not the breakdown of determinism but rather of our simplistic models of it.

4. Biff on March 6, 2012 1:45 PM writes...

@1 Rick Wobbe - Investing in biotech is not investing. It's gambling. Call it "speculating," if you're looking for a more polite term. Aside from a very small number of cases, the industry has not been kind at all to those who put their money into it.

5. Travis on March 6, 2012 1:51 PM writes...

"There's not a lot of algorithmic compressibility in this business."

What about automation? Can the empirical data gathering be automated in a way to quickly try millions, billions, or trillions of permutations?

I'm thinking "cloud computing" reduced to "cloud biochemistry". But then again... I'm a roboticist / electrical engineer, so I have no clue about the subtleties involved.

6. Biff on March 6, 2012 2:13 PM writes...

In a previous life, I worked on microtubules. (I'm a card carrying cell biologist, by training.) One of the finest scientific presentations I ever attended was a physics department seminar, given by a proper physicist who had decided to abandon the world of frictionless surfaces to investigate the physics of biological systems, in this case, the capability of microtubules to perform mechanical work by distorting a lipid vesicle.

This was quite a few years ago, so I'm a little hazy on the scientific details and the identity of the physicist, but I will never forget the moment when he flashed on screen a set of equations familiar to anyone with high school physics, and then he proceeded to add a slew of terms to the equations, terms that physicists normally skip or cancel out by assuming things like frictionless surfaces, perfect elasticity, and perfect vacuums. The screen rapidly became cluttered with mathematical expressions, and the audience of physicists became visibly agitated. The introduction of "solution viscosity" as a major factor seemed to cause at least one aneurysm. I heard a member of the audience remark later that it was "madness" to study such things.

Madness, indeed. I loved the physicists' reactions almost as much as the science itself.

7. HelicalZz on March 6, 2012 2:21 PM writes...

And there we have one of the big underlying issues that needs to be faced by the hardware engineers, software programmers, and others who come in asking why we can't be as productive as they are.

Well for one thing, drugs rarely have the ability to 'add new features' to the system in which they work (such as with enzyme replacement therapy). Few can even 'activate' malfunctioning processes. The vast majority inhibit existing processes in some shape or form, with the intent that such interference will have a net positive outcome. It is inherently difficult in any abstract system to improve it by reducing its ability function, but most drugs attempt to do just that. Is it any wonder that unintended consequences can crop up with this kind of 'tinkering'. The approach can make sense in systems gone haywire, such as with cancers, but diseases due to poor function are difficult to affect in this manner. It is a testament in my opinion that so much has been accomplished with such a limited tool set.

Zz

8. William B Swift on March 6, 2012 2:26 PM writes...

Reductionism and determinism are supreme in all interactions above the quantum level. "Emergence" is simply an admission that the interactions between the factors are too complicated for our minds and computers to model and forecast in anything approaching real time.

You might find the essays indexed on this page of the LessWrong Wiki helpful, http://wiki.lesswrong.com/wiki/Reductionism

9. biologist on March 6, 2012 4:13 PM writes...

If you talk to a knowledgeable computer scientist about drug discovery, you can just say: drug discovery is an NP-HARD problem, i.e. worse than NP-complete. The traveling salesman problem (find best route for a traveling salesman to visit a number of cities, i.e. shortest way or shortest time) is an NP-complete problem. The computation time for this quickly grow beyond any boundary if you look at a larger number of cities. More see Wikipedia.

10. Emmanuel Gustin on March 6, 2012 5:18 PM writes...

Speaking as a phycisist who somehow got involved in drug discovery, I would be inclined to argue exactly the opposite position: Many of the standard assumptions in drug discovery are extremely reductionist, to a level that most physicists would be highly uncomfortable with. Yes, many people in the field are very aware of the high complexity of biological systems, but that doesn't mean that they have found effective ways to deal with it. There are great efforts being made in systems biology and related branches, but these still have to enter mainstream drug discovery.

It is true enough that physics often resorts to simplification of the models, because our mathematical ability to solve a system of equations is enormously inferior to our ability to write down complex equations involving many factors. Therefore physicists often drop the smaller terms, not because they like to, but because they have to. Fortunately, progressive refinement of the approximations often works. As is demonstrated by the chemistry of anything beyond the hydrogen atom.

But at least this approach produces the habit of considering several factors, ranking them in order of importance, and designing the best approximative model, valid under certain assumptions. Many (not all) areas of drug discovery still have to progress to this stage: There is no way yet to get a handle on all the important factors, and ranking them is often impossible. The alternative is often enough guesswork, and that may be one factor contributing to the dismal success rate of drug discovery projects.

I can understand that that many people with a background in biology or chemistry are reluctant to embrace the kind of mathematical modeling physicists use, because it seems to simplify so much. But the alternative is not necessarily better. The mental assumptions people make instead are often equivalent to very simple linear models, or even to binary boolean logic. I think it is possible to use to use the tools of physics and mathematics to generate models that are better than that -- still far from perfect, but better.

Of course these models would have a speculative element, but that isn't any different in physics. Theoretical predictions still need to be tested in the lab, to see which of the possible models is right.

11. leftscienceawhileago on March 7, 2012 2:59 AM writes...

Derek,
I am concerned that you are giving a bit of a misleading impression about AIT...

We don't know if there is or isn't a lot of "algorithmic compressibility" in this business, the compressibility is uncomputable!

12. Georg-Martin Krapper on March 7, 2012 3:37 AM writes...

Given the state of data analysis in Medicinal Chemistry, it will be a while yet before Drug Discovery becomes something that physicists will identify with (or even recognize as science). Here's a comment (that actually got published in a journal) asserting the superiority of logD7.4 over logP for solubility prediction.

"The clearer stepped differentiation within the bands is apparent when logDpH7.4 rather than logP is used, which reflects the conisderable [sic] contribution of ionization to solubility.”

13. SS on March 7, 2012 9:11 AM writes...

Derek,

After having run the MRL internal scientific libraries and IT groups before the Equinox layoffs of 2003, and having interacted with pharma executives at numerous pharmas, I can tell you that in my humble opinion, the reaction of most pharma execs and beancounters (esp. those lacking rigorous scientific backgrounds) to the issues in this post could be described as the Bovine Stare of Incomprehension.

They would neither understand the reductionistic view nor the wicked-problem view.

They would just want to know why drug discovery is not like building new iPhones.

14. Anon on March 7, 2012 9:29 AM writes...

@2. Nekekami
I'd love to know more about the mathematics involved in "serious psychology" as you call it. If you could link to an example that would be much appreciated.

15. Rick Wobbe on March 7, 2012 9:46 AM writes...

@Emmanuel Gustin, #10,
I take your point that the problem may not be with the applicability of reductionism/determinism per se, but rather the attendant problem of simplification. However, there's a small part of me that wonders, for reasons I can't articulate, if reductionism "may not even work in principle" as Curious Wavefunction suggests. It is a philosophically arguable (and well-argued) point...

With regard to your comments about the superiority of mathematical models, however, I must challenge you to provide experimental evidence based on carefully defined studies. People seem to generally accept that this is true, but I haven't seen any published studies supporting such claims in the case of drug discovery.

I'd also like to suggest that you raise a false choice that gets raised too often, perhaps unintentionally: that the choice between mathematical models and human reasoning/induction/intuition is a mutually exclusive one. Perhaps it follows from the inherent binary nature of financing decisions (i.e. funding one option excludes funding others), but industry-wide adoption of one to the exclusion of the other seems foolish. It seems wise to bear in mind George Box's famous admonition, "All models are wrong, some are useful"

16. RKN on March 7, 2012 12:42 PM writes...

I presented at a molecular medicine conference recently on the topic of network biology, discussing our integrative algorithms for identifying candidate biomarkers and companion diagnostics. Earlier, I sat in on a talk in a parallel track listening to the chair of the session give his spiel. His opening slide posed this question:

"What do we need more of in medicine, doctors or algorithms."

Being a systems biologist I lean toward the latter.

17. Nekekami on March 7, 2012 1:26 PM writes...

@14, I'm not a psychologist myself, nor do I have a degree in it, I'm basing it on what a friend who has studied it told me, and the basic courses I received as a NCO. Therefore I can't give you any exact references, but there's a whole field called mathematical psychology which, if I understand things correctly, is used a lot on the research side of things(but not on the counseling side), and has taken in a lot of input from various fields of physics.

The problem, as I got it explained for me, is that, while in physics, you can simplify and work with an "ideal" problem, to keep it elegant and easy to work with, in psychology you work with real problems in real populations, and all those factors a physicist would throw away as "irrelevant" can in fact have impact

18. Anonymous on March 7, 2012 2:42 PM writes...

@Rick, #15,

Yes, it is an interesting philosophical point whether there might be systems in biology that are resistant to simplification and approximation. One can argue that, for example in chaos theory, such systems also exist in classical physics. It this time, the answer has to be speculative.

Still, if we consider what biological evolution is likely to do, I personally wouldn't expect to find too many systems that are truly irreducible. And I don't mean that evolution has to favour "simple" solutions; that is clearly not the case. There is nothing to stop organisms to develop intricate pathways full of positive and negative feedback loops, obsolete branches, superfluous links between different systems, and redundant pathways. Quite the contrary, probably. The only simplifying influence may be the genomic tendency to re-use existing mechanisms for other purposes, which may lead to a lot of repetition of fixed patterns.

But evolution is also likely, I expect, to drive organisms towards relatively stable solutions and to limit the number of inputs a system is sensitive to. There isn't a discernable advantage in having a metabolism that can go haywire at random moments. And signalling pathways that are unrelated in purpose can still be interlinked, but the response to extraneous signals has to be somehow damped down, or it would become a fantastic nuisance. Of course, cancer cells and auto-immunize diseases demonstrate the imperfections in this: Obviously, there still is a potential in the systems to short-circuit or break down. But those are, fortunately, the exceptions and not the rule.

So my guess is that when we understand the mechanism better, we may find that cells have developed excessively complex mechanisms to achieve relatively simple results, because these mechanisms were evolved and not designed. And that therefore, a reductionist approach may well be applicable. Not always and not for all purposes, but often.

And I would agree that the intuition of a scientist and an attempt to predict outcomes by mathematical modeling can be in agreement; there does not have to be a dichotomy or a conflict there. However, the intuition of a scientist is the product of his training and experience. Therefore there may very well be a case for giving drug discovery biologists some basic training in the modeling of dynamic systems, even if they never use it, to give their intuition a broader basis. I am not saying that correct intuition cannot be achieved in other ways; but even a limited experience in modeling dynamic systems can adjust one's intuition a lot.

As for practical examples, that is an interesting question. It doesn't have to be complex; I once did some simple modeling of dose-response curves in a relatively simple cellular assay, and found that effects that had been routinely dismissed as experimental errors had in fact a mechanistic basis.

But my favorite illustration of how signalling pathways can lead to non-intuitive results is still rather theoretical: Tyson et al., Current Opinion in Cell Biology 2003, 15:221–231. The paper does include references to a number of practical examples. I don't think many people could look only at the signalling diagrams and "intuitively" sketch the corresponding signal-response curves.

19. Emmanuel Gustin on March 7, 2012 2:43 PM writes...

@Rick, #15,

Yes, it is an interesting philosophical point whether there might be systems in biology that are resistant to simplification and approximation. One can argue that, for example in chaos theory, such systems also exist in classical physics. It this time, the answer has to be speculative.

Still, if we consider what biological evolution is likely to do, I personally wouldn't expect to find too many systems that are truly irreducible. And I don't mean that evolution has to favour "simple" solutions; that is clearly not the case. There is nothing to stop organisms to develop intricate pathways full of positive and negative feedback loops, obsolete branches, superfluous links between different systems, and redundant pathways. Quite the contrary, probably. The only simplifying influence may be the genomic tendency to re-use existing mechanisms for other purposes, which may lead to a lot of repetition of fixed patterns.

But evolution is also likely, I expect, to drive organisms towards relatively stable solutions and to limit the number of inputs a system is sensitive to. There isn't a discernable advantage in having a metabolism that can go haywire at random moments. And signalling pathways that are unrelated in purpose can still be interlinked, but the response to extraneous signals has to be somehow damped down, or it would become a fantastic nuisance. Of course, cancer cells and auto-immunize diseases demonstrate the imperfections in this: Obviously, there still is a potential in the systems to short-circuit or break down. But those are, fortunately, the exceptions and not the rule.

So my guess is that when we understand the mechanism better, we may find that cells have developed excessively complex mechanisms to achieve relatively simple results, because these mechanisms were evolved and not designed. And that therefore, a reductionist approach may well be applicable. Not always and not for all purposes, but often.

And I would agree that the intuition of a scientist and an attempt to predict outcomes by mathematical modeling can be in agreement; there does not have to be a dichotomy or a conflict there. However, the intuition of a scientist is the product of his training and experience. Therefore there may very well be a case for giving drug discovery biologists some basic training in the modeling of dynamic systems, even if they never use it, to give their intuition a broader basis. I am not saying that correct intuition cannot be achieved in other ways; but even a limited experience in modeling dynamic systems can adjust one's intuition a lot.

As for practical examples, that is an interesting question. It doesn't have to be complex; I once did some simple modeling of dose-response curves in a relatively simple cellular assay, and found that effects that had been routinely dismissed as experimental errors had in fact a mechanistic basis.

But my favorite illustration of how signalling pathways can lead to non-intuitive results is still rather theoretical: Tyson et al., Current Opinion in Cell Biology 2003, 15:221–231. The paper does include references to a number of practical examples. I don't think many people could look only at the signalling diagrams and "intuitively" sketch the corresponding signal-response curves.

20. Rick Wobbe on March 7, 2012 4:20 PM writes...

Nicely said, Emmanuel (#18 & 19). Very Tao. If we meet in another life, it'd be fun to work together.

21. your_majesty on March 7, 2012 10:45 PM writes...

#5: Interesting thought but I don't think it will work. Chemical space is way, way too large and drug binding is equally too nuanced for any luck here. To get an idea, just try computing the number of molecules encoded by the average drug patent. It is easy to think of a few trillion trillions and be tricked that this is close to infinity.

22. Luke Weston on March 8, 2012 6:31 AM writes...

I think this is, maybe, how at least some physicists think they might solve the world's drug discovery challenges:

Step 1: Build a scalable quantum computer.

Step 2: Input and solve the equations of quantum mechanics for a complete QM description (in polynomial time) of the particular biochemical target molecule you're interested in, and generate lead compound structures from that.