Skip to main content

Drug Development

Tone It Down

You know, I’m not sure I should have cut Atomwise so much slack the other day. I just came across this piece on them, and. . .well, I’ll let it speak for itself:

“Here I am just sitting in this house and I’m able to predict a cure to measles,” co-founder of Atomwise Alex Levy tells me over the phone from his apartment in Mountain View, Calif.
Atomwise, a health tech startup in the current Y Combinator batch, has launched more than a dozen projects in the last year to find cures for both common and orphan diseases – diseases that would otherwise be too expensive and time-intensive to tackle. It’s working with IBM to find a cure to Ebola and with Dalhousie University in Canada to search for a measles treatment. Levy says the startup went through 8.2 million compounds to find potential cures for multiple sclerosis in a matter of days.
. . .“It’s like having a virtual super intelligent brain that can analyze millions of molecules and potential interactions in days instead of years,” Levy says. . .“You still have to test but you take out all the guesswork before you get started.”

OK, guys, this is what’s known as hubris. That’s a term from the ancient Greeks, and it translates to something like “overconfident pride” in modern usage, and it’s a good way to summon the goddess Nemesis. We get to meet her a lot in this business. Retribution lands on us several days a week around here in drug research, as you will note from our success rates in the clinic. But this sort of “Gosh, I’m curing diseases right here at my kitchen table” talk is pushing it even further, inviting a visit from another ancient Greek chick, the goddess Atë. She’s in charge of delusional folly, and pitches in on the retribution as well. Atë’s the one that Shakespeare has helping out Caesar’s spirit in Mark Antony’s speech, you know, cry “Havoc” and let slip the dogs of war and all that stuff.
What I’m getting at here is that this interview is the kind of publicity you don’t need. The sorts of people that you’re going to have to work with to get a drug project off the ground are going to roll their eyes (at best) in reaction to it, and it’s going to be hard to get anyone to take you seriously if you go on in this vein. This work is hard enough as it is, without making it even harder.
Update, from the comments: “Alex Levy here, the fellow quoted in the piece, and longtime Pipeline reader. To Derek and most commenters, thanks for the critique. Simply put: we agree. We’re always looking for better ways to communicate about Atomwise, and computational drug discovery generally. It’s challenging to help the press and public differentiate hits from cures, and evidence from proof. What a system like Atomwise actually does can be pretty opaque to a general audience.
We’re skeptical scientists too, and do extensive retrospective and prospective validation studies internally. We hope to publish more as time goes on, and are excited to share those results with the community here. Wouldn’t it be nice if Atomwise works even half as well as it sounds on TechCrunch?”

63 comments on “Tone It Down”

  1. adam says:

    You’re going to see this a lot from YCombinator-funded companies. It’s Silicon Valley culture applied to problems which aren’t easily solved by pretending you can save the world…when they fail, they’re going to complain about The Man (or Big Pharma) keeping them down.

  2. Anonymous says:

    “…it’s going to be hard to get anyone to take you seriously if you go on in this vein.”
    But isn’t this kind of grandiose talk the sort of thing that brings in the investment money? If that’s the current goal, then maybe it’s well planned strategy.

  3. gogoosh says:

    Well, in their defense, he doesn’t say he’s able to *correctly* predict a cure for measles. Heck, I’m able to predict the final score of Super Bowl LXXXVII from my apartment.
    Seriously, though, you take out ALL the guesswork?

  4. Biotech Capitalist says:

    ok, the latitude I think early stage companies deserve to make bold statements does have limits. And it looks like Atom-no-so-wise crossed into the unrealistic.
    If only they promised to double or triple hit rates in HTS, that itself would be a great accomplishment.

  5. John Wayne says:

    They are bragging about a virtual screen of 8.2 million compounds? Either they don’t get the concept of chemical space, or they are running their software on an iPhone.

  6. The Iron Chemist says:

    Oooh… “Potential cures,” those are as good as actual cures, right?

  7. Anon the modeller says:

    And to think the CEO wants med chemists to take in silico more seriously. To the chemists, if a modeler cannot tell you where the limitations of a modeling are, and clearly point to the potential sources of error (of which there always are some), you cannot make an informed decision on the results. A good modeler will always confess to his ignorance and unsureness. We share your pain!

  8. luysii says:

    It’s touching to see such faith in our increasingly secular world

  9. #1 has it nailed.
    This kind of hubris is common in the tech scene in Silicon Valley. “Come and change the world!” screams the billboard for a web browser company.
    The difference is that Silicon Valley used to steer clear of biotech. Not anymore! The dollars are just sloshing around on the tech side, so now they are spilling over into other industries like biotech.
    Expect to hear a lot more of this in the future.
    Honestly I think it’s a bit undignified. The one thing I like about scientists is that they are naturally skeptical. “Oh you think X causes Y? I can list 10 reasons why that’s not true.” That is slowly going away as more of the tech culture infiltrates the biotech culture.
    Anyone else thinks this smells like the late 1990’s promising of genomics? We all know how that ended: a ten year drought in biotech investing as people finally figured out most of it was hype.

  10. anon the II says:

    I’ve remarked on this blog often about molecular modeling. My complaint has generally been, it’s not the modeling that’s so bad, it’s the people doing it.
    They’re not all bad, but most are. Thanks for another data point.

  11. Aaron says:

    There’s a bit of technical discussion here:
    Basically, they’re assuming that neural networks with the recent developments from Hinton et al will work substantially better than current in silico methods. This seems foolish to me, as recent NN method gains rely on enormous training sets. They think they can use all previous chemical results as a training set, but you also need an enormous training set of *hits*. If we already had a list of 1000+ compounds that bind a target effectively, we wouldn’t need the neural network in the first place.

  12. RD says:

    We already have a cure for measles. It’s called an MMR vaccine. Why is he reinventing the wheel?
    As a modeler, I resemble those remarks about modeling. I try to be upfront about limitations. If only med chemists would be equally upfront about claiming they got all their ideas.

  13. Anonymous says:

    I think it’s charitable of you to characterize this as hubris. It’s downright stupid.
    It reflects such a magnitude of ignorance and arrogance that it’s hard to believe this person isn’t deliberately misleading investors.

  14. AndrewD says:

    Now perhaps I don’t understand this modelling stuff, but I thought the idea was to generate structures which should have biological ativity. The molecules will still need synthesising by the Derek’s of this world and then we Process development chemists will need to produce robust manufacturing processes neither of which are likely to be easy or economic. Have the modellers considered this?

  15. Grim Reaper says:

    Hubris is being kind to this crowd.- “The first principle is that you must not fool yourself – and you are the easiest person to fool.” – R. P. Feynman.

  16. Anon says:

    Could you please not refer to a woman (even mythological) as a “chick”? Thanks.

  17. DCRogers says:

    There’s a lot of excitement about so-called “Deep Learning” right now – and honestly, I think it is a true advance (even if it takes a HORRID amount of computational power).
    That noted, my bias in HTS-style campaigns has always been really-fast-and-good-enough: strip-mine 90% of the information quickly, the results will be close enough to the ‘best’ results to be useful to the chemist (or not very useful, at which point one would question whether all the work to get 99.6% of the information included would have changed anything).
    There has been a flurry of new startups based around this technology, and I wish them well. For some problems, I think it will provide a real advantage. For this problem, I just don’t see an enormous trove of yet-unmined information in our screening data (though I could be wrong, and so look forward to them testing this out).
    Now apply this technology to solving one of the super-hard problems in chemistry (binding energy estimation, protein dynamics, structure-to-function) and I start to get interested…

  18. watcher says:

    How about disclosure of their successes, drugs in the clinic as a start, approved by regulators as the follow-up. If that comes into play we should listen. anything less is just false bravado by someone who doesn’t know the real purpose….not just talking big to grab a bunch of unearned cash.

  19. Anonymous says:

    There is a great quote from the trailer for the new season of HBO’s silicon valley.
    The CEO of the rival to the startup says “I don’t want to live in a world where someone else makes the world a better place, better than we do.”

  20. Awesome sauce says:

    Roll it up, light it up, smoke it up
    Inhale, exhale

  21. RD says:

    @14 Andrew:
    Oh, where is the snark tag when you need it?
    Some would say that modelers rarely consider whether their designs have biological activity. Others say that modelers rarely consider whether their designs are within the realm of reality, what with the occasional pentavalent carbon. Let’s just say that modelers are frequently misunderstood.
    I never put together a design that couldn’t be synthesized. Sure, there might have been too many chiral carbons for comfort or the bond might have been unstable. But if some med chemist put a similar substructure in a database, I’m going to check it out.
    But yeah, the topic has come up before.

  22. anonomoujs says:

    Totally agree with the tenor of the discussion here, but for me this is only marginally more offensive than the current commercial for IBM’s Cloud in which the researcher states with similar pride (hubris?) that he is ‘computing a cure for cancer’.

  23. anon the II says:

    I’ll paraphrase:
    Computational chemistry, the cure of the future, and always will be.

  24. MoBio says:

    Reminds me of a conversation I had some years ago with a Boston-based VC where they said they were intrigued in the platform but were taking a pass on it because:
    “It couldn’t predict side-effects that we didn’t know about”

  25. Chrispy says:

    I have come to wonder about the necessity of delusion in breakthrough science — “visionary” people I have known are frequently delusional. The ones who do change the world, though, typically have a way of getting people to buy into their delusion. You have to sound credible. Levy sounds like he doesn’t know what he’s talking about.

  26. DB says:

    @2 – it clearly got them some Y Combinator money, but that’s not getting anything through clinical trials. They need pharma money, and they need clinicians who will take them seriously and work with them.

  27. Anonymous says:

    That’s the problem i saw looking though the list. Most if not all of the companies require somebody else to still be investing in bringing treatments to market.

  28. bad wolf says:

    Hubris, orphan diseases, high tech pie in the sky? Can’t believe Ethan Perlstein doesn’t have an interview in the same issue.

  29. Ted says:

    These guys aren’t dumb. They’ll pull in cash to fund their development work, and they’ll promise the moon and stars in exchange for modest royalties. The great thing is, they’ll sit in the project background, completely divorced from the reality of making the compounds, screening the hits, etc… As the projects die off, each will be written off as unique exceptions. Of course, each will be with a different partner, so their ‘industry validation’ will continue to grow, even as they fail to deliver.
    By the time the partner stream dries up, they’ll have spent the better part of a decade working their way through the industry, sucking up and spending a big pile of money and well-positioned to move on to the next great computational problem (like ‘supercomputer designed optimum diets’).
    Their model doesn’t require successful results, they just need successful partnerships.
    Of course, a blind squirrel finds the occasional nut, so they could always win the lottery, and find the “cure for measles”.

  30. Iamthep says:

    What is your opinion of the results of this Kaggle competition:
    I think most here are underestimating the effect that this technology will have on biology.

  31. bad wolf says:

    @12–sorry no, Silicon Valley/Bay Area families do need a cure for measles since they’re notoriously undervaccinated (link in handle). “Too smart” or something.

  32. steve says:

    I second #16 – there really is no reason to refer to a woman as a “chick”. We should be past that already.

  33. anonymous says:

    Derek, I forwarded your previous post to someone who had read that blurb about atomwise to give them an idea of what to believe. I am so happy to forward this one to them 😉

  34. MoMo says:

    You have to believe Atomwise because I cured syphilis once while in my bathrobe, while watching The View and having my nails painted.
    Actually, he should be darted with Combelen then tagged for territorial range and behavioral studies. A species like this only comes across once in a lifetime and you have to study them quickly.

  35. Anonymous from 2. again says:

    That’s kind of my point. I doubt that they are really after pharma money. Their goal is to get other investors to give them money for awhile, and not really to cure measles (or anything).

  36. Anonymous says:

    That sort of talk calls for a GSK acquisition….

  37. HT says:

    @Derek – Hubris, more hubris, blasphemy, and Atomwise?
    @30 – if by effect you mean that this’ll detract administrators and investors from supporting research that *might* have **some** chance of a positive outcome for biomedicine … then yes, we are probably underestimating it

  38. luysii says:

    How many of the commenters here would have uttered similar sentiments about the synthesis machine a few years ago?

  39. Ann O Mouse says:

    You know, I think we should cut these guys a break. They are unabashed and unapologetic in their hubris and the reasons ($$) behind it and they are clearly excited about their method. They call it predicting a cure for the splash value and because that’s something anyone can appreciate. It looks to me like what they are predicting is target affinity for lots of compounds faster (better?) than anyone else. I hope they succeed. Clearly they and anyone with any knowledge understands that they are trying to improve the odds and timeline for finding a drug, not feeding lots of data in one end and getting THE structure of the cure out the other. They know the risks and how far we are from finding a cure in silico. I would feel differently if they had, oh I dunno, published a paper in science about their technique and then talked about how we were on the verge of finding cures for everything via calculation alone, and lots of Pharma scientists joined in the chorus. Let’s give them a chance to be fall flat, but hope they actually win big.

  40. Ann O Mouse says:

    See my comment # 40. There is no synthesis machine. Just a machine that can carry out one type of reaction. That was hubris under the guise of science. This is just plain hubris, no disguise. I prefer the latter.

  41. Anonymous says:

    Deep learning is really a hype atm.
    Even in image prediction (for which they are known) they have their pitfalls e.g. ;

  42. Ted says:

    I like this tidbit from the TechCrunch article:
    “Widespread use of antibiotics and antibacterial products have helped diseases like pneumonia become resistant to the drugs.”

  43. Anon2 says:

    When I was a child I remember our school having a fundraiser to cure cancer. I didn’t really know what cancer was (and I definitely didn’t know there were multiple types), but they were telling us week after week to sell items out of a catalog to raise money. I believe the grand prize was a ride a stretch limo to a pizza restaurant (come to think of it, I haven’t seen a stretch limo in a while). Anyways, they kept telling us “we are only $xxxx away from a cure!” That reminds me a lot of what is going on here. You have a person/group who thinks they know where the goal line is and if they could only do task Y they will have a cure. It just doesn’t work that way when there is so much we don’t know about diseases.
    I also applaud Derek for pumping the brakes and doing what he can as an individual to prevent outlandish statements/actions from potentially discrediting the problem. We’ve really got to respect the problem if we want to address it appropriately.

  44. Ryan says:

    Deep learning is for sparse datasets. The set of all images 256×256 pixels is vast. But the ratio of interesting images to all possible images is very very small. This is what I mean by sparse.
    The same thing applies to molecular biology. The number of conformations that a protein could take is vast, but we are only interested in a few. Just like the number of possible proteins is vast, but we are only interested in a few.
    Deep learning’s hype is well deserved. It is definitely something that people should be paying attention to.

  45. Anon says:

    Nothing is wrong with calling a goddess “chick” here (especially on his personal website). Internet police don’t agree with me though.

  46. Ted says:

    Chill out, dude.
    Oh, wait. Am I allowed to say dude anymore?

  47. David says:

    @45 Images are a great example. Deep learning for computer vision is pretty great until it gets tricked by things it won’t normally see.
    How does this relate to molecules? Well, the cornell authors optimized the images to the DNN. If you optimize a molecule using your DNN, you likely will end up with similar crap.

  48. Esteban says:

    I hear young women use ‘chick’ all the time and not with the intent of being derogatory. It really is just the female equivalent of ‘dude’. Stay strong Derek.

  49. bad wolf says:

    Right on, Derek. Don’t let Steve and rest of the anonymous PC Grammar Nazis tell you what to write!
    Classical goddesses offended by offhand remarks can submit their own responses.

  50. diver dude says:

    Yet again, I have a problem with all this snark. Atomwise are proposing a solution for a problem we have been complaining about for years – low productivity in drug development (by which I mean very few compounds making it to actually treating patients).
    It’s a hypothesis so let’s see them test it. Something needs to change in our business and pointing and laughing is not going to do it.
    I think #25 has it right. And Arthur C. Clarke put it this way, “If an elderly but distinguished scientist says that something is possible he is almost certainly right, but if he says that it is impossible he is very probably wrong.”

  51. Alexander Levy says:

    Alex Levy here, the fellow quoted in the piece, and longtime Pipeline reader. To Derek and most commenters, thanks for the critique. Simply put: we agree. We’re always looking for better ways to communicate about Atomwise, and computational drug discovery generally. It’s challenging to help the press and public differentiate hits from cures, and evidence from proof. What a system like Atomwise actually does can be pretty opaque to a general audience.
    We’re skeptical scientists too, and do extensive retrospective and prospective validation studies internally. We hope to publish more as time goes on, and are excited to share those results with the community here. Wouldn’t it be nice if Atomwise works even half as well as it sounds on TechCrunch?

  52. featherson says:

    @16-Shut up asshole.

  53. Anonymous says:

    @52: Could you perhaps provide more details on how your approach works, in particular how it’s different from what’s come before and why it’s better? I understand you’re using deep learning, but that’s typically applied to things like images which have fixed-length representations. How do you apply it to something like a protein structure or drug molecule? More details would help dampen the skepticism.

  54. Questioner says:

    @52 Glad to hear that you feel you’re misunderstood by Pipeline readers. How about giving us a paragraph or two written for an informed audience about what you think you’re going to do better than all the other guys?
    Differentiating between IDing screening candidates for lead optimization vs coming up with marketable drugs would be a good start, because right now your press releases seem to be conflating the two.

  55. John-john says:

    Re 52: “We’re skeptical scientists too*”
    * and staggering unrepentant dirt-merchants playing buzzword bingo for the big payoff.
    Sorry, but what you are proposing has absolutely no basis in any reality- ‘it’s not even wrong.’ It’s just smoke, mirrors and manure.

  56. DrSnowboard says:

    never mind the quality, feel the marketing…

  57. HoodMonkey says:

    I thought chick was acceptable slang. When did the broads decide it wasn’t?

  58. Esteban says:

    @58: It’s not so much the broads as it is the dames who are offended.

  59. MoMo says:

    Did you know that the word “Chicks” started the PC movement in the mid 1980’s in Philly? Dr. Tom Lynch said the word in a medical lecture about the limbic system, where he stated “What do you think alligators think about all day? Eating, excreting, and cruising the bars for Chicks. Next thing you know the female students got huffy, the papers picked it up, Lynch becomes the agent provocateur for his freedom of speech attacks, and the rest is history.
    You Chicks just don’t get it. If I was an alligator I’d cruise the bars for chicks as well. Its just part of their diet.

  60. Marvin says:

    Here I am, brain the size of a planet, and they ask me to take you to the bridge. Call that job satisfaction, ’cause I don’t.

  61. Marcela says:

    For those who may have missed the Shakespeare reference:
    Marcus Antonius:
    And Caesar’s spirit, raging for revenge,
    With Ate by his side come hot from hell,
    Shall in these confines with a monarch’s voice
    Cry “Havoc!” and let slip the dogs of war,
    That this foul deed shall smell above the earth
    With carrion men, groaning for burial.
    Julius Caesar Act 3, scene 1, 270–275

Comments are closed.