About this Author
DBL%20Hendrix%20small.png College chemistry, 1983

Derek Lowe The 2002 Model

Dbl%20new%20portrait%20B%26W.png After 10 years of blogging. . .

Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases. To contact Derek email him directly: Twitter: Dereklowe

Chemistry and Drug Data: Drugbank
Chempedia Lab
Synthetic Pages
Organic Chemistry Portal
Not Voodoo

Chemistry and Pharma Blogs:
Org Prep Daily
The Haystack
A New Merck, Reviewed
Liberal Arts Chemistry
Electron Pusher
All Things Metathesis
C&E News Blogs
Chemiotics II
Chemical Space
Noel O'Blog
In Vivo Blog
Terra Sigilatta
BBSRC/Douglas Kell
Realizations in Biostatistics
ChemSpider Blog
Organic Chem - Education & Industry
Pharma Strategy Blog
No Name No Slogan
Practical Fragments
The Curious Wavefunction
Natural Product Man
Fragment Literature
Chemistry World Blog
Synthetic Nature
Chemistry Blog
Synthesizing Ideas
Eye on FDA
Chemical Forums
Symyx Blog
Sceptical Chymist
Lamentations on Chemistry
Computational Organic Chemistry
Mining Drugs
Henry Rzepa

Science Blogs and News:
Bad Science
The Loom
Uncertain Principles
Fierce Biotech
Blogs for Industry
Omics! Omics!
Young Female Scientist
Notional Slurry
Nobel Intent
SciTech Daily
Science Blog
Gene Expression (I)
Gene Expression (II)
Adventures in Ethics and Science
Transterrestrial Musings
Slashdot Science
Cosmic Variance
Biology News Net

Medical Blogs
DB's Medical Rants
Science-Based Medicine
Respectful Insolence
Diabetes Mine

Economics and Business
Marginal Revolution
The Volokh Conspiracy
Knowledge Problem

Politics / Current Events
Virginia Postrel
Belmont Club
Mickey Kaus

Belles Lettres
Uncouth Reflections
Arts and Letters Daily
In the Pipeline: Don't miss Derek Lowe's excellent commentary on drug discovery and the pharma industry in general at In the Pipeline

In the Pipeline

« Uranium, Eh? | Main | Pharma Overview »

December 9, 2011

Drugs, Airplanes, and Radios

Email This Entry

Posted by Derek

Wavefunction has a good post in response to this article, which speculates "If we designed airplanes the way we design drugs. . ." I think the original article is worth reading, but some - perhaps many - of its points are arguable. For example:

Every drug that fails in a clinical trial or after it reaches the market due to some adverse effect was “bad” from the day it was first drawn by the chemist. State-of-the-art in silico structure–property prediction tools are not yet able to predict every possible toxicity for new molecular structures, but they are able to predict many of them with good enough accuracy to eliminate many poor molecules prior to synthesis. This process can be done on large chemical libraries in very little time. Why would anyone design, synthesize, and test molecules that are clearly problematic, when so many others are available that can also hit the target? It would be like aerospace companies making and testing every possible rocket motor design rather than running the simulations that would have told them ahead of time that disaster or failure to meet performance specifications was inevitable for most of them.

This particular argument mixes up several important points which should remain separate. Would these simulations have predicted those adverse-effect failures the author mentions? Can they do so now, ex post facto? That would be a very useful piece of information, but in its absence I can't help but wonder if the tools he's talking about would have cheerfully passed Vioxx, or torcetrapib, or the other big failures of recent years. Another question to ask is how many currently successful drugs these tox simulations would have killed off - any numbers there?

The whole essay recalls Lazebnik's famous paper "Can A Biologist Fix A Radio?" (PDF). This is an excellent place to start if you want to explore what I've called the Andy Grove Fallacy. Lazebnik's not having any of the reasons I give for it being a fallacy - for example:

A related argument is that engineering approaches are not applicable to cells because these little wonders are fundamentally different from objects studied by engineers. What is so special about cells is not usually specified, but it is implied that real biologists feel the difference. I consider this argument as a sign of what I call the urea syndrome because of the shock that the scientific community had two hundred years ago after learning that urea can be synthesized by a chemist from inorganic materials. It was assumed that organic chemicals could only be produced by a vital force present in living organisms. Perhaps, when we describe signal transduction pathways properly, we would realize that their similarity to the radio is not superficial. . .

That paper goes on to call for biology to come up with some sort of formal language and notation to describe biochemical systems, something that would facilitate learning and discovery in the same way as circuit diagrams and the like. And that's a really interesting proposal on several levels: would that help? Is it even possible? If so, where to even start? Engineers, like the two authors of the papers I've quoted from, tend to answer "Yes", "Certainly", and "Start anywhere, because it's got to be more useful than what you people have to work with now". But I'm still not convinced.

I've talked about my reasons for this before, but let me add another one: algorithmic complexity. Fields more closely based on physics can take advantage of what's been called "the unreasonable effectiveness" of mathematics. And mathematics, and the principles of physics that can be stated in that form, give an amazingly compact and efficient description of the physical world. Maxwell's equations are a perfect example: there's classical electromagnetism for you, wrapped up into a beautiful little sculpture.

But biological systems are harder to reduce - much harder. There are so many nonlinear effects, so many crazy little things that can add up to so much more than you'd ever think. Here's an example - I've been writing about this problem for years now. It's very hard to imagine compressing these things into a formalism, at least not one that would be useful enough to save anyone time or effort.

That doesn't mean it isn't worth trying. Just the fact that I have trouble picturing something doesn't mean it can't exist, that's for sure. And I'd definitely like to be wrong about this one. But where to begin?

Comments (36) + TrackBacks (0) | Category: Drug Development | Drug Industry History


1. Wile E. Coytoe on December 9, 2011 12:32 PM writes...

Perfect example is estrogen. What genes does it turn on? what genes does it turn off? How does it do each of those? What are the consequences of turning on gene A, but not gene B, or instead inhibiting gene B? Estrogen does one thing in cells in tje uterus, others in the breast, others in the vagina, other in the brain, other in the pituiary.... Not all cells in those tissues react the same; for example uterine myometrial cells vs uterine endometrial stroma vs uterine endometrial epithelium. Not all reactions are the same depending on progesterone, prolactin, and other hormone levels. Humans don't respond to estrogen exactly like rats, mice or dogs, so what are the species specific responses in each cell and tissue? I'd really appreciate it if someone could draw me a diagram that takes all of that into account so that I can fully understand the biology here. Looks like a radio to me.

Permalink to Comment

2. Anonymous on December 9, 2011 12:59 PM writes...

ugh, this, again. Why is it so damn hard for those (supposedly) sentient engineers to understand ? For a toaster, or a computer, or a plane, you can draw a blueprint because you already know all the parts that are there, and what they do. For an organ, or a cell, or a nucleus, you can't, because - oh I give up...

Permalink to Comment

3. John Wayne on December 9, 2011 1:29 PM writes...

I think it would be a great idea to try and come up with a unified way to present biological systems. The critical word in the last sentence is 'try.' Challenges will include raw complexity, nonlinear effects, and high order dependence. Also, thirty to forty percent of the published results are incorrect; more will be irrelevant to in vivo systems.

It seems like we're in an analogy war with people who haven't experienced contemporary drug discovery process first hand.

"It would be like aerospace companies making and testing every possible rocket motor design rather than running the simulations that would have told them ahead of time that disaster or failure to meet performance specifications was inevitable for most of them."

Drug discovery is just like designing a new rocket with the following caveats: (1) nobody has ever made a rocket that can go that fast (new biochemical target), (2) some rocket designs always have inherent safety flaws that have to do with the destination rather than the rocket design itself (on mechanism toxicity), (3) current predictive models are somewhat good at predicting old rocket designs but not new ones (need new basis sets), (4) your boss wants a new working rocket in two quarters (or you're fired), (5) at least thirty percent of the literature on building rockets is wrong (probably higher than 30), and (6) we don't actually know how current rockets work (CNS indications).

Permalink to Comment

4. Sigivald on December 9, 2011 1:57 PM writes...

would that help? Is it even possible? If so, where to even start? Engineers, like the two authors of the papers I've quoted from, tend to answer "Yes", "Certainly", and "Start anywhere, because it's got to be more useful than what you people have to work with now". But I'm still not convinced.

I can't see how to disagree with any of those answers.

They might well be underestimating the complexity factor, but it's not like it's going to be easier to deal with a really complex biological system by not having shorthand notations and the like to deal with it.

(After all, biology is chemistry and chemistry is physics. It's just a lot [a WHOLE lot] messier and interconnected than single-particle or two-atom physics simulations, as Mr. Wayne says.

But you have to start somewhere, to get anywhere, and while the scale and complexity difficulties are Very Serious, there's nothing involved that makes it fundamentally un-modelable or un-predictable.

It is "just physics", in the end. The only problem is that the physicists don't realize that it's so damned much physics when they make the suggestion.

["If it was EASY everyone would be doing it."])

Permalink to Comment

5. RM on December 9, 2011 2:02 PM writes...

these little wonders are fundamentally different from objects studied by engineers. What is so special about cells is not usually specified,

A radio is designed. Each part is there because it serves a purpose. Moreso, it is there because it is the simplest (or cheapest) part there is which will serve that purpose. It was also likely chosen from a standard parts catalog, and conveniently color coded or labeled. There ain't no epicycles in radio design. The human brain can handle only so much complexity, so anything it designs by necessity has to comprehensible to the human brain.

In contrast, a cell is evolved. (Now) useless parts tend to stick around unless they're costly to produce. (Imagine if your iPod still had a useless AM radio demodulator in it, because no one got around to removing it yet.) Old parts get reused for completely new purposes. (This resistor isn't functioning as a resistor - it's actually just a mechanical support for this part over here). Everything's a one-off, and tuned to the system it's in.

I can't remember where I read it, but I remember a story about researchers who used a programmable processor (an FPGA) to evolve a circuit. The final design only had a small part of the chip electrically connected to the input and output, but removing unconnected parts rendered it non-functional. The chip relied on low-level electromagnetic fluctuations between the used and "unused" parts of the chip to function - fluctuations which aren't captured by the standard equations and modeling of electrical circuit design.

Permalink to Comment

6. Frank Adrian on December 9, 2011 2:35 PM writes...

I'll admit that I'm not a chemist - just an computer engineer. The furthest I went in chemistry was second semester inorganic. Even so, through my normal perusal of blogs, popular science reading, and other anecdotal evidence, I understand that biological systems are orders of magnitude more messy than even the worst systems ever designed by man (In fact, it's probably the main argument against intelligent design - there's very little that looks designed, if you actually look at it). Some foolish engineer's notion that drug discovery is constrained by systematization issues and not biology is , frankly, insane.

Computational chemistry is still not at a point where you can get protein conformations and solvation effects right, let alone large-scale interactions. And without proper computer simulation, you're dead in the water as far as modern engineering techniques go.

It really just boggles my mind that anyone who took the time to look at anything having to do with chemistry and/or drug discovery would spout ignorant stuff like this. They just have no clue. More importantly, they're dangerous because they minimize the amount of work that needs to actually be done and make others think that more can be done more quickly with less - something that doesn't seem to be true at this time.

Permalink to Comment

7. ex-Pfizerite on December 9, 2011 2:43 PM writes...

In effect we do model, every time we put an experimental drug into a rat, mouse or dog we are modeling the effect of that drug on a human. The attempt to model a drug in-silico is more than 7 orders of magnitude more difficult than an aerospace design effort for a new plane. I would even argue that a clinical trial is an in-vivo modelling effort in that we look at the effects of the drug on a target population. The problem is that drug research is not engineering but basic science even when the drug goes out into clinical trials and continues even as the drug is used in the clinic.

Can a computer model tell me a priory why do humans and guinea pigs require vitamin C and penicillin is toxic to guinea pigs? Then explain the other idiosyncratic effects of drugs between different species.

Permalink to Comment

8. OldLabRat on December 9, 2011 2:59 PM writes...

Engineers I know get nervous when the data matrix for building simulation models is "only" 85% full. Bio pathway modelers would be in raptures if 15% of the data matrix was known. Perhaps the Andy Groves of the world might focus on filling the data matrix, or at least understanding how sparse the available modeling data is.

At least a rational biology ontology/nomenclature would be a great help and needs to start somewhere.

Permalink to Comment

9. Sundowner on December 9, 2011 3:03 PM writes...

I cannot understand why is so hard to understand that designing something for the scratch is completely different from understanding something other people has designed. And that engineer approach to the problem does not help: tell an engineer that he has to do reverse-engineering on the technology made by a crazy alien over 1.000.000 years using not design principles but trial and error and Rube Golberg machines.

So drug discovery for SNC makes real sense !!! We need a drug targeted to the receptors of common sense. Commonsensine???

In a crazy way it reminds me a comment by Charles Sheffield about physicists gathering in a congress by the end of the XIX Century; they reached the conclusion that in physics most things were already explained, but two small, annoying problems: radiation of a black body and the speed of the light... and we know what those two small minor questions rendered in the end.

So thinking that med chems and biologist are just a bunch of stupids that have all the tools at hand but they simply they do not know how to use them, seems so arrogant...

Permalink to Comment

10. Biotechtranslated on December 9, 2011 3:11 PM writes...

Great comments so far, especially #1 and #5.

I love the analogy of an iPod with an AM receiver in it, but let's take that further...

The best analogy I can give the engineers is imagine a laptop computer that was built with not just the latest technology, but all the electronic technology developed over the last 100 years (vacuum tubes, incandescent lights, spark-gap transmitters up to modern ICs). Now the computer works just fine and is comparable to the latest and greatest. Now you want to upgrade it.

Well, you can't actually open the computer, you're just allowed to look in one of the vents on the side. So you jam your screw driver in their and pull out a vacuum tube and replace it with a transistor. Turn it back on and it doesn't work. Well, what you didn't know is the thing generates so much heat that the transistor burned out. Ok, so you replace it with the original vacuum tube. Back to square one. So now you replace spark-gap transmitter with a modern antenna. Oops, it doesn't work now because the weird frequencies from the spark-gap transmitter that were there, and were attenuating another interfering frequency, are gone and now another component is failing. Rinse and repeat.

Now take that computer and multiply the complexity by several magnitudes and throw in some components you've never seen before.

Welcome to drug discovery!


Permalink to Comment

11. Joe Corkery on December 9, 2011 3:18 PM writes...

I find the following quote to be particularly troubling:

"Every drug that fails in a clinical trial or after it reaches the market due to some adverse effect was “bad” from the day it was first drawn by the chemist."

This presupposes that there is such a thing as a drug without adverse effects and ignores the fact that toxicity is relative in importance when compared to the therapeutic function. Oncology drugs are allowed to have much broader adverse effect profiles, while obesity drugs are not.

He also ignores the fact that clinical trial design is often as an important factor in failure as the compound itself.

Permalink to Comment

12. NoDrugsNoJobs on December 9, 2011 3:31 PM writes...

I hate this kind of crap though it does remind me to be more cautious before I spout off about subjects outside my own expertise.

Permalink to Comment

13. DCRogers on December 9, 2011 3:37 PM writes...

> Why would anyone design, synthesize, and test molecules that are clearly problematic, when so many others are available that can also hit the target?

I'm assuming this is rhetorical, as it has a why-don't-you-stop-beating-your-wife format that cannot be answered as-is.

There's an ongoing debate as to how much weight to give simulation evidence, but to promote it to the status of oracle that provides 'clear' evidence is just as dumb as the implication that chemists routinely weight that evidence to zero.

Permalink to Comment

14. Brooks Moses on December 9, 2011 3:49 PM writes...

So, I'm an engineer....

I did my Ph.D. research on some theory related to design of fuel injectors for diesel engines. The current state of the art in designing fuel injector nozzles is that you can do simulations of part of the flow, and make some educated guesses if it's an evolutionary change to an existing design, but when it comes right down to it, the only way to tell if a given fuel injector nozzle design will actually produce an evenly-distributed spray of appropriately fine droplets is to test it.

(John Wayne's description in comment 3 about "what if designing a rocket was like this" is really not that far off for fuel injectors. Among other things, it is surprisingly difficult to see what's going on in the middle of a dense cloud of droplets with non-intrusive methods.)

And a fuel injector is significantly less complicated than a multicellular organism.

One thing that hasn't been mentioned, though, is that the success conditions are significantly more complicated for biological systems than for many mechanical ones. For an airplane, the goal is clear -- the lift force on the plane is equal to the weight. And that's something that immediately reduces to basic principles; you compute the lift force on each piece and sum them up, and the weights and sum those up, and there you go. We can draw an arbitrary box around the plane and integrate within it -- and so, if you tell me that you've got a plane (or a bird, or superman!) in a box and the airflow going in has a given velocity, I can use those integrals to tell you things about the airflow going out.

Biology isn't like that. What does "5% decrease in mortality from type-XYZ lung cancer" mean at that sort of level?

Permalink to Comment

15. John Wayne on December 9, 2011 3:57 PM writes...

I’ve been thinking about this topic, and this sentence seems appears to have two critical assumptions:

“Why would anyone design, synthesize, and test molecules that are clearly problematic, when so many others are available that can also hit the target?”

1. Medicinal chemists do not usually ‘design, synthesize and test molecules that are clearly problematic’. We avoid common pitfalls (reactive and promiscuous functional groups, etc.), and use in silico, in vitro and in vivo modeling of toxicity on every project. People disagree on if this is a good method based on the known disconnect between all this stuff and humans; we may just be spending a lot of money and time optimizing something that is still a die roll.

2. There are not ‘so many other molecules that can also hit the target’. I have never, not even once, been a part of a project that had enough good leads to follow up on. We check our relatively small number of potential compounds out and work on the one or two we figure are the best. This subject gets in to the value of compound sample collections; the classic example is GSK screening 80 novel microbial targets and getting no hits on most of them.

Permalink to Comment

16. RKN on December 9, 2011 4:46 PM writes...

There is a branch of system's biology that tries to model molecular activities in a cell using differential equations, linear or non-linear depending on the problem. I appreciate the efforts of these people but it's very difficult, not so much because the math is hard, but rather because it is difficult to obtain reliable, quantitative readouts of the relevant inputs and outputs (e.g., metabolic products) necessary to value these equations. So far as I know we're not even close to doing this for a single signal transduction pathway in one normal cell, to say nothing of understanding it in a perturbed state or multiple cell types.

Worse, the biological phenotypes we're most in understanding in are not likely to be elucidated merely by compiling a comprehensive parts list of the cell. Just like a detailed understanding of the individual parts in a plane won't give you its stall characteristics, for example.

Permalink to Comment

17. Anonymous BMS Researcher on December 9, 2011 11:36 PM writes...

I can speak with authority on the deep differences between engineering and biology because I once was an engineer before getting my doctorate in biology and ending up in pharmaceutical research. Engineers work very hard to make their designs (1) understandable and (2) modular. By modular, I mean the interface between one component and the rest of the system is made as simple as possible. Also in many cases we have standards to define interfaces, precisely so everybody can stop thinking about them. Back in the 1970s my father was a Visiting Professor in England. At that time, there were about eight different types of electric outlets in older houses. They had recently defined a standard for new construction, so today most places in the UK have that type. But back then, appliances were sold with bafe wires coming out of the cord so the customer could install the right plug for tge outlets in his or her house. Our local Woolworth store had one aisle dedicated to adaptors for making various plugs fit various sockets. Now, none of that is needed unless you live in an old house that hasn't been rewired in 40 years. Most people can just buy gadgets, take them home, and plug the in.

Every aspect of biology is totally non-modular and non-standard. The same part can interact with many other parts in a zillion ways. Engineers don't do that. Engineers want to have each component only interact with certain other parts in specific ways, so that it is possible to design a complex system without being overwhelmed by its complexity. The person who designs the water pump does not need to know all about the rest of the car; he or she is told a few specifications for that pump fits into the engine then goes off and designs it.

Permalink to Comment

18. MIMD on December 9, 2011 11:40 PM writes...

As both a physician/medical informatics specialist and a ham radio enthusiast, extra class, I'll say it simply:

The 'Syndrome of Inappropriate Confidence in Computing' runs rampant in techies.

Permalink to Comment

19. Dan on December 10, 2011 12:13 AM writes...

I'm in engineering, but I'm well aware it ain't that easy. You have to have an accurate (and reasonably simple, otherwise running the simulations will take longer than just making the stuff and testing the real thing) model of how everything interacts for it to be useful in any way. For integrated circuits, we have that sort of model, refined by decades of work in the field (the basic models that they teach you in intro-to-electronics break down once you get into the 200nm regime). For biology, we're not even close yet.

It's difficult enough to understand an existing chip design if you have the source HDL code and access to the designer to get your questions answered. If you want to consider the human body in the same terms, consider that: 1) it's an infinitely more complex system, 2) at best, we have the equivalent of a gate-level netlist (characterization of various proteins) and at worst the physical layout and nothing more (DNA), and 3) the human body doesn't look all that "designed" - and if there is a designer, he's not answering questions.

Permalink to Comment

20. Sundowner on December 10, 2011 3:48 AM writes...

Now that I think about it, this example will be appreciated by engineers. If they have read 'The Mote in God's Eye' by Niven and Pournelle, just think about the Moties approach to engineering vs. the Human approach to engineering. The human engineers in the book are flabbergasted when they see how the Moties' technology works.

It is hard science fiction, so they should like it. Let's read it, think about it and come back for more.

Permalink to Comment

21. Vince on December 10, 2011 6:04 AM writes...

Dan, I greatly appreciate your comment, but I'd contend it's still missing the point. Why would we need a low-level description on the same level of abstraction as a netlist or VHDL?

Again, you're not thinking like an engineer. What matters is the I/O mappings given your function (ie. small molecule, nanoparticle, etc). To think of anything else is to get caught up in weeds and forget to keep your eye on the ball: which, sadly, even many of the self-professed engineers seem to have done here.

Stop thinking like a biologist and this is coming from someone trained in neurobiology. Remember, the vast majority of the information content in any system is contained at the lowest levels; it's also generally of little utility.

Just as the chemists here don't seem very concerned with the goings-on down at the Planck length when designing their molecules, we can achieve much without a full description.

This shouldn't come as a revelation, given every second generation anti-psychotic or anti-depressant compound in existence was designed under conditions in which we have sparse knowledge about the system, little knowledge about the underlying pathology, no firm grasp on what to design for -- yet are designing molecules which have measurable I/O effects on arbitrary tests like the HAM-D. This is SOP for us.

Does Seroquel, which the ad which tells me is an antidepressent, work primarily through NET inhibition as claimed? What about Effexor and Cymbalta? But what about the