Skip to Content

Alzheimer's Disease

The Case of Verge Genomics

A number of people have passed along the recent press stories about Verge Genomics, a new company out of YCombinator that has just raised $32 million for neuroscience drug discovery. Now that, as literally anyone who’s ever done it can tell you, is a hard field of a hard field, and I wish Verge good fortune in getting something to work. There’s a horrifying amount of what we always call “unmet medical need” in that area, and anything that actually goes toward alleviating it is welcome. So again, good luck to them.

But Verge does not, it seems, believe that any luck will be necessary. They are perhaps the purest confluence yet of several trends in biopharma startups. First is the AI/machine learning field, which I’ve written about several times and most certainly will write about again. Second, unfortunately, is  “What the stodgy old drug industry needs is some happenin’ young disruptors”. That one comes up as people look at how long it takes to develop a drug, how much money it costs, and how many blind alleys get explored, and decide that the time is ripe to be the Amazon of drug discovery (upending the yawn-sleepy world of retail shopping), the Craigslist that demolished newspaper classified ads, the Apple that triumphed with the iPhone (but definitely not the Apple that almost went under in the 1990s).

Note that all of these examples contain plenty of semiconductors and lines of code. That’s another related trend, which we might call “Silicon Valley To The Rescue”. I’ve written about that a number of times here as well, of course, as regards the Andy Grove Fallacy (biomedical advances should take place at the speed of computing hardware advances), which is a subset of a general confusion about the physical world bred by experience bringing new hardware and software to market. If you can create a new device with new functions and have it adopted by millions of people as part of their everyday lives, you have indeed accomplished something. But that accomplishment might give you an exaggerated idea of your overall ability to affect reality as well.

That’s not because fabricating new chip designs and writing good code are easy: no, those things are hard. But drug discovery and medical advances, sadly, are even harder. The track record so far for the Silicon Valley/drug discovery interface is mixed (followup on that last link). So when I see things like this from a startup like Verge, I wonder:

(Alice) Zhang was three months shy of her MD and PhD graduation from University of California-Los Angeles when she left school to start Verge Genomics in 2015 with Jason Chen, who she met during the program.

“I just became very frustrated with the drug discovery process,” she said. “It’s largely a guessing game where companies are essentially brute force screening millions of drugs just to stumble across a single new drug that works.”

At the time, Zhang also recognized the advancements in neuroscience, machine learning and genomics occurring all around her. Genome sequencing had become more and more affordable, and breakthroughs in understanding how function connects with genes opened a new field of possibilities for exploring disease and health. And there was an opening for an opportunity to (take the) guesswork out of drug discovery. The vision for Verge was to become the first pharmaceutical company that automated its drug discovery engine, helping to rapidly develop multiple lifesaving treatments in diseases like Alzheimer’s disease, ALS, and Parkinson’s disease where no cure exists today.

As you read the rest of the article (and others like it) you come statements about how Verge is using human data “from day one” (unlike those drug companies), and how they’re not just looking at one gene at a time. And then there’s this one “Instead of tediously screening millions of drugs, the algorithm will computationally predict drugs that work“.

Something about that sounds vaguely familiar. Hold on. . .here we are:

. . .(these computer techniques) make drug design more rational. How? By helping scientists learn what is necessary, on the molecular level, to cure the body, then enabling them to tailor-make a drug to do the job… This whole approach is helping us avoid the blind alleys before we even step into the lab…  Pharmaceutical firms are familiar with those alleys. Out of every 8,000 compounds the companies screen for medicinal use, only one reaches the market. The computer should help lower those odds … This means that chemists will not be tied up for weeks, sometimes months, painstakingly assembling test drugs that a computer could show to have little chance of working. The potential saving to the pharmaceutical industry: millions of dollars and thousands of man-hours.

That’s the ticket! And that, as some readers will recognize, is from an infamous article from 1981. Soon that piece will have its fortieth anniversary; we should have some sort of celebration. And yes, I understand that one of the whole points of a company like Verge is that it’s not 1981 any more, and that we have both vast amounts of data, and the abilities to deal with vast amounts of data, that no one could have gotten their heads around back then. But what if those still aren’t addressing the rate-limiting steps?


Ash Jogalekar summed it up perfectly on Twitter yesterday, as shown at right. The problems with Alzheimer’s, ALS, and Parkinson’s drug discovery are not data handling problems. The important problems with drug discovery in general are not data handling problems, and unfortunately there are many people who would like to think that they are. Who would perhaps like to think that everything could be solved if we could just obtain and correlate enough data. But what we’re short of is insights, ideas, and understanding, and those come slowly, painfully, and expensively.

Even taking Verge’s approach on its own terms, I very much doubt that we – we humans – have the tools to connect functions with genes in quite the way that the publicity would indicate. Nor do I think that these diseases are necessarily best approached from a pure genomic direction. That’s a part of the puzzle, but it’s a mighty big puzzle. Nor do I think that there are algorithms yet that will take long lists of compounds, correlate them with genomics data, and predict the winners. The Verge folks may have been misquoted, or had their claims exaggerated along the way by the press coverage. And Verge can come and prove me wrong about that, and in a way I hope that they do. But as it stands, unfortunately I think that the only people who can wholeheartedly believe in this approach to drug discovery are people who have never done any drug discovery.

Well, the article says that they have six drugs in development, “closer to the clinical end”, for what that’s worth. Let’s see what happens, and how long it takes, and how much money it costs. That’s the great thing about this field; we occasionally settle questions. Does this compound work? Is this target valid? Do we have a useful drug? My own experience is that we can indeed get those answers, but that it can be a long and viciously expensive process. Over to you, Verge – come on down and take a crack at it, and good luck to you.


118 comments on “The Case of Verge Genomics”

  1. Anon says:

    Let me know who their investors are, so that I can go to them with my idea for a perpetual motion machine to power faster-than-light travel.

    1. Uncle Al says:

      It’s been done (Hollywood script), re the Casimir effect executed in volume not inverse distance…Hyperbolom, LLC spacetime gate. “Beyond vacuum to…nothing”

      Perpetual motion is easier. Conservation of mass-energy derives from Noether’s theorems plus time homogeneity. Time is not homogeneous within a gravitational well (re GPS). All that remains is reduction to practice. Get the investors’ names. We’ll incorporate in Azerbaijan.

      Pharma is harda.

      1. Aleksei Besogonov says:

        Fun fact: energy (and impulse) is not conserved in the General Relativity. You can describe metrics that satisfy the field equations and yet allow generating any amount of power.

        So yes, let’s start signing infestors!

  2. Terry says:

    Elizabeth Holmes redux. Prove me wrong.

    1. Hap says:

      Why assume evil when arrogance, incompetence, or stupidity will do almost as well? If the founders have a history of dishonesty, then evil becomes more likely, but without that sort of evidence, it seems unreasonable to assume evil.

      1. Nick K says:

        I think wishful thinking on the part of the founders and investors is a more likely explanation.

    2. Mad Chemist says:

      That’s exactly what I was thinking. Hopefully, Ms. Zhang learns some humility before following that path too far.

    3. Dave Howe says:

      I see there is a book now – – So at least someone is getting something out of the Theranos debacle 🙂

  3. TMS says:

    They have one published patent application mainly to a compound which has a wikipedia entry: isoflupredone.

    “Isoflupredone, also known as deltafludrocortisone, as well as 9α-fluoroprednisolone, is a synthetic glucocorticoid corticosteroid which was never marketed.”

    1. mikey says:

      can you put a link to this published patent application – I couldn’t find any.

      If you can’t put the link can you put your search strategy – I searched for the assignee as “verge genomics” (nada!). what did you search for to pull it up?

      1. Anonymous says:

        United States Patent Application, 20170360805
        Inventors: Chen; Jason; Zhang; Alice
        Applicant: Verge Analytics, Inc., San Francisco, CA

        1. mikey says:

          thanks mate. appreciated.

  4. Anon says:

    “The problems with Alzheimer’s, ALS, and Parkinson’s drug discovery are not data handling problems.”

    Well they kind of are: We don’t have any positive data to handle in the first place!

    No algorithm can ever make a positive prediction if we’ve never observed a positive result!

    This is much more than a misunderstanding of biology or drug development, it’s a misunderstanding of basic statistics, computing and information theory, which is supposed to be their core expertise!

    1. eub says:

      Yeah, this.

      Backing up — machine learning models are trying to learn a function from input to output, generalizing from ‘golden’ (input, output) training data. (There are are other kinds of machine learning but the point applies.)

      What’s their output, clinical success? For a new application there’s only training data (SomeCompound, FAIL), there’s no SUCCESS output to be had. If you step backwards and train for surrogate metrics, then you need to understand the biology to bet that your surrogates will pan out for you.

      This ain’t rocket science but it seems like a problem for them.

      The broader thing is, pharma has serious “Big Data” only where pharma can mechanize the scaling of work — Big Data of compound bindings, of assays, of anything with a microarray in it, maybe there can be bacteria. The data about biological outcomes in whole animals is more expensive and so smaller. And the data about clinical outcomes is tiny data in machine learning terms.

      1. eub says:

        Rule of thumb: you can build a plan around machine learning to behave as well as a human expert, but don’t plan on it being superhuman. People can recognize faces, now computers can, that’s great. People can’t look at the existing pharmacological body of knowledge and predict reliably what compounds will be active at a target and what will work clinically. I mean, people try, and you can plan on replicating that level of guesswork in software (sorry, good luck everybody in that job market), but you should not plan on eliminating the guesswork.

    2. SearchMasterFlex says:

      CLEARLY it’s not a data problem, it’s a SEARCH problem. If you had an infinite number of monkeys on typewriters typing nonsense and then you were to just Google the right words, a cure would pop up-up. QED.

  5. Anon says:

    It also gives me great confidence in this outfit that they tried to recruit me (BA chemist in the industry for ~10 yrs) for their Head of Medicinal Chemistry position. Can’t say it wasn’t flattering!

    1. Scratching Heads says:

      They had that posting for their Head of Chemistry for more than a year now. It’ll be interesting to see if they will eventual hire anyone and who will accept that responsibility.

      1. Bagger Vance says:

        It’s kind of amusing to compare the requirements for that job to Ms Zhang’s own credentials. Of course, this position would report to CSO, who then reports to Ms Zhang(?)

        “Ph.D. in organic or medicinal chemistry” vs “dropped out of MD-PhD”
        “Minimum 10 years of industry experience” vs “I just became very frustrated with the drug discovery process”
        “Track record of SAR success” vs “We’re rethinking how drug discovery is done”

  6. Emjeff says:

    Why in the world would you drop out of school 3 months shy of graduation? That does not impress me; in fact, it makes me think Alice Zhang is impulsive.

    1. Anon says:

      Agreed. Maybe she’ll decide to go back into college “3 months shy of delivering a cure for Alzheimer’s”.

      In other words, this seems more like a cop out from her expectation of failing to graduate, than a rational, well-thought decision to make a real sacrifice.

    2. Billy says:

      She probably wanted to join the ranks of Bill Gates and Larry Ellison. What’s the point of finishing school when you can convince VCs to give you money and dream about membership in the tres commas club?
      This reminds me, I gotta go buy a lottery ticket.

    3. Dr Zoidberg says:

      I wondered the same thing. My first thought was that she didn’t leave but was on the verge of failing, though I backed up and wondered if she was maybe completely burned out. Neither would be good for the founder of a company though, so it doesn’t really matter. I don’t find that line impressive either, but I think Elizabeth Holmes has changed the stigma on leaving school early (yes, despite her fall from grace).

      1. Paula says:

        Ooooh That’s it! On the VERGE of failing… That’s why they named the company “VERGE” LOL

    4. Anon, PhD says:

      I was in grad school with her. She had at least 2-3 years left. Note she has no first author publications (and wasn’t even close to one). All the exaggeration from Verge is intentional.

      1. Gregr says:

        No first author pubs and 2-3 years from graduation (at the the very least) is a huge difference from 3 months that was claimed. It didn’t make sense to me either. This is not an experienced independent researcher.

        1. anon says:

          “was three months shy of her MD and PhD graduation from University of California-Los Angeles when she left school to start Verge Genomics”
          Inspired by Theranos?

      2. pete says:

        interesting. Thanks for sharing that.

        And don’t tell her investors ’cause it’ may be greeted as “Way too much information !”.

    5. zmil says:

      Sounds like someone didn’t want to write their dissertation…

    6. michf says:

      Whatever the case with Alice Zhang, the rest of their team seems very legit. The co-founder appears to be an actual MD-PhD with dozens of high quality publications and patents (most of them first author) in many different fields; the rest of their team has first author publications in Cell, Nature, Science; their advisory board is a who’s who of famous biologists in the field. I wouldn’t discount this company so out of hand with literally zero supporting arguments besides “she is a woman and so is Elizabeth Holmes”

      1. flies says:

        ‘zero supporting arguments besides “she is a woman and so is Elizabeth Holmes”’
        eh what? The argument you’re responding directly to is, “she says she was 3 months from graduation, but her publication record makes that appear dubious.” Also, there’s this whole article somebody wrote about the problems with naive tech startups in pharma, hold on I’ll find the link…

      2. zero says:

        More like “this person is making exaggerated claims and taking in huge investments on near-zero evidence, much like Holmes. Holmes was a fraud, as are most ‘innovators’ or ‘disruptors’ in pharma / med development, especially those who say they will ‘revolutionize’ something with ‘AI’. It is reasonable to suspect that this person might be a fraud as well until we see evidence otherwise.”

        Note the lack of gender signifiers in that statement. This has nothing do do with the pronoun and everything to do with the content resembling a pattern of silicon-valley bullshit.

      3. YouHaveNoIndustryExperience says:

        You are an idiot. They have academic credentials which are not the same as industry drug development experience. Writing papers does not translate into doing good SAR for drug optimization. Just because you can swing a baseball bat doesn’t mean you will be a good golfer

  7. Dear Prof Lowe,

    I continue to be dismayed by your lack of open-mindedness with respect to Artificial Intelligence (AI), a field which will undoubtedly disrupt and transform medicine beyond what your limited imagination can conceive. Did you know, for example, that AI can enable the direct observation of experimental non-observables (such as Ligand Efficiency) and that quantum coherence transfer can give water molecules long term memory. The key design parameter in neuroscience is to scale the head capacity at constant volume by the product of the dipole moment and the trace of reduced hyperpolarizability tensor.

    Yours sincerely,

    Kígyó Olaj
    Program Leader, Budapest Enthalpomics Group (BEG)

    1. Dolores Abernathy says:

      “Have you ever seen anything, so full of splendour?”

      1. anon says:

        These violent delights have violent ends.

    2. Anon says:


      AI is really nothing more than multiple regression, so how can any algorithm make a reliable positive prediction if we’ve never observed a positive result?

      Making predictions is easy, but you can’t reliably predict what you’ve never observed. That’s basic math and information theory, not just faith or opinion.

      1. Kígyó Olaj says:

        Dear Dr. Anon,

        You have clearly been reading too many of these blog posts and have become infected by the same negativity that afflicts Prof Lowe so grievously. Limitations of the multiple regression techniques, in which you have such Faith, include a need to account for numbers of parameters and the requirement that models be validated. These problems evaporate when disruptive, machine learning tools are applied and, as a bonus, correlations between parameters can simply be ignored.

        Kígyó Olaj
        Program Leader, Budapest Enthalpomics Group (BEG)

        1. Anon says:

          Regardless, AI is pattern recognition, and you can’t recognize patterns and thus make and test predictions if you never observe them in the first place.

          Can you give me 3 examples of AI-based predictions that predicted something that was never observed before? In other words, which worked outside and beyond the scope of the given data set?

          1. Anon says:

            … actually I’ll give you a week to post just 1 example for others to judge, after which we can safely assume you’re selling AI snake oil.

          2. b says:

            (It’s satire)

          3. Peter Kenny says:

            Hungarian snake oil even…

          4. Anon says:

            Then I open up this challenge to *anyone* working in “AI”. Because I’m sick of hearing inflated expectations that I know cannot possibly be true for fundamental reasons.

          5. anon the II says:

            So many things to say.

            To Hap: it’s called Hanlon’s razor (

            To Anon arguing with Dr. Olaj: Argue with a fool and he’ll do the same. Actually, I think Olaj was tongue in cheek.

            To BiotechFanatic: I think we saw yesterday in Finland, just what you get with “naïveté combined with hubris” will get you.

            That’s all for now.

          6. Dolores Abernathy says:

            From the Hungarian website:
            “The essence of this novel approach is that, in addition to to its contributions to enthalpy and entropy of binding, each molecular interaction will now be awarded points for the artistic elements of the contact between ligand and target. This industry-leading application of Big Data uses the Blofeld-Auric Normalized Zeta Artificial Intelligence (BANZAI) algorithm to score aesthetic aspects of molecular interactions”.

            As Dr. Ford told me, if the Blofeld-Auric doesn’t give it away (see: Bond, James) then the BANZAI (see Buckaroo) should.

          7. Hap says:

            Yes, it’s not original, but it did seem valid.

          8. Chris Phoenix says:

            Look at AlphaGo Zero: It learned to play top-rank Go, even by modern computer standards, with zero human input.

            I don’t think AI is just multiple regression anymore. Some of it is actual evolutionary discovery.

          9. Progress has been made says:

            Siri and Alexa understand novel speech every day. The spoken phrases weren’t in the training data, and neither were the speakers.

            Now, I expect a skeptic’s first objection is that Siri was trained to understand speech. Yet Verge would still be useful if it merely worked for the general class of proteins and molecules.

            A skeptic’s second objection would be that Siri and Alexa misunderstand some phrases. Yet humans are not perfect in all cases. I don’t think any drug hunter claims 100% success rates.

          10. Derek Lowe says:

            Saying “if it merely worked for the general class of proteins and molecules” is a phrase that needs some more explanation, I fear.

    3. me says:

      The other commentors in this post clearly are not well read in their Enthalpomics.

      1. Dr. Demented says:

        Chris Phoenix: AlphaGo Zero is a remarkable achievement. However, a significant difference between Go and drug discovery is that the rules are definitively known in Go, and not even close to being known for drug discovery – particularly for neuroscience drug discovery where Verge has chosen to play.

        1. zero says:

          Precisely; a game of Go has a definite end and a specific result (win, lose, tie), not to mention very simple rules. The learning algorithm was given the opportunity to observe many games with a variety of results. This is exactly the opportunity that a ‘drug discovery expert system’ will be denied.

    4. Earl Boebert says:

      When I worked for Honeywell in the 1980s we applied a variety of what were then called “AI techniques” to a variety of military problems such as target recognition. Our marketing motto was “If it works we call it pattern recognition.”

    5. John Wayne says:

      I’d love to see a tensor that described the motion of a peptide (the simplest possible thing to model)

    6. Anonymous says:

      Ligand Efficiency, etc.: There is so much more to drug discovery than just binding. There are as many or MORE negatives (tox; selectivity; off-target effects; PKPD; target mutations; drug interactions; compensating pathways; …) that determine success.

      As far as binding is concerned, one of my favorite pre-computation examples is methotrexate (Subbarao, 1947) which was predicted by humans to inhibit DHFR by mimicking folic acid and binding to that site. It does bind to and inhibit DHFR but at a completely different binding site, as shown decades later by x-ray. Despite the incorrect prediction, MTX remains an important and effective drug. Also, it binds to the IL1-beta receptor and probably other targets.

      Irwin “Tack” Kuntz’s DOCK program successfully predicted many good binders in silico. That included some that bound strongly in vitro but, by x-ray, were bound in a completely different, other-than-predicted fashion.

      I don’t care about the upside-down and backwards predictions if the experimental data actually lead to a drug.

      Derek had a recent Pipeline discussion of “The Entropic Term is Laughing At Us By Derek Lowe 11 July, 2018”

    7. Duane says:

      Now that is some 4-D chess ninja level trolling. Bravo.

  8. GumpyC says:

    In a world where one the the Kardashians is a “self-made” billionaire, what’s the harm in a little self promotion in the pharma startup scene?

    1. Chrispy says:

      No, this kind of gee-whiz disruptive garbage makes it harder to raise money for real approaches that might actually work.

  9. Earl Boebert says:

    I would suggest that any discussion of the application of the Silicon Valley model to other fields (which model is, today, a hairsbreadth away from pump-and-dump) must consider the emphasis that model places on being early or first to market. That emphasis has the (unstated) side effect that the company relies on its customers to discover its mistakes. Get Rev 0 out there, take the revenue, upgrade to 0.1, rinse and repeat. Very cool if the product is a smartphone game, not so cool if it holds human life in its hands.

  10. tlp says:

    From their website: “we’ve discovered a way to map out the hundreds of genes that cause a disease, and then find drugs that target all the genes at once”

    Is it how you reframe phenotypic screening for Ycombinator to give money?

    1. Gregr says:

      And how in the world to target hundreds of genes simultaneously with a single small molecular inhibitor? It takes medicinal chemists years to optimize a drug for a single target.

      1. tlp says:

        Pretty much any molecule will affect hundreds of genes if you have an instrument sensitive enough to detect the change.
        Their approach seems not very different from high-resolution imaging ML described in this blog few months ago. There’s a reasonable chance they’ll discover some clinical candidate eventually. Probably not for Alzheimer’s, Parkinson’s or ALS but something good enough to sell to bigger pharma for profit.

        1. Gregr says:

          Yes, that’s probably what the promo material from Verge means when it says that they are targeting hundreds of genes at the same time. Targeting indirectly, but not directly. So then this is not so different than what big pharma does: target a single protein that then has cascading effects on the expression levels of other genes.

  11. ChrisP says:

    You should check insitro from Daphne Koller for a company that got VC money with little to show so far. For Verge, if true, they have assets and are 3 years old now (young but not just a week old before larger VC investment). Good luck to them.

    1. Joe says:

      At the very least, Daphne Koller has some experience, connections, and credentials to justify a blind-faith investment in her company, which is no sure bet either. But you can at least understand funding a person with her track record in other pursuits.

  12. BiotechFanatic says:

    The sad piece here is that naïveté combined with hubris divert resources away from work that could help address some of the actual bottlenecks in neurodegenerative work. Admittedly the prognosis for those $ spent isn’t particularly good either, but surely we’d be in with a chance of actually contributing to the foundation of knowledge on which Polyanna’s neurodegenerative pharma armamentarium will be built.

  13. Jack says:

    I have been reading your blog for years…thanks for sharing so many valuable insights over that time.

    When it comes to predicting drug targets via algorithmic methodologies, the future is already here…Compugen has developed a predictive discovery platform powered by proprietary algorithms and will have a wholly owned asset in the clinic(COM701) in the next three months. It also has a partnered asset(BAY1905254) going into the clinic before year end.

    Both of these are completely novel discoveries.

    COM701 is the one to watch…there is a strong biological rationale, and if, and yes that’s a BIG if, it translates in the clinic some large PD-1 franchises will scramble to control it.

    Btw, 701 was developed in conjunction with Johns Hopkins, and the work on this molecule there was led by none other than Drew Pardoll (you’ll find his name on all of the 701 abstracts) who also chairs the CGEN scientific advisory board, which is impressive in and of itself. Drew also sits on board of US subsidiary.

    Of course these assets have to prove themselves in human clinical trials, but the future may be closer than anyone thinks.

    Here’s their corporate presentation:


    1. Anon says:

      Good luck, but then if AI is so good and reliable, you won’t need it, right?

      I’m guessing you didn’t remortgage your house and use your kid’s college fund to maximize your equity stake in this sure-fire bet.

      And that’s the real issue here: AI is only “guaranteed” to work when it’s somebody else’s money on the line!

      1. Jack says:

        Who said anything about a guarantee?

    2. MrXYZ says:

      Two targets in (or almost in) clinical trials is obviously a good place to be. The question is, did the computational methods get you there anymore efficiently than the standard discovery methods? How many targets turned out to be false leads?

      Having asked that, I do like Compugen as a company. I know that there is a lot of experimental validation of the targets that come out of their computational methods.

      As in all things, trust but verify!

      1. Jack says:

        I believe that their method is faster and cheaper than industry standard, and that they certainly have had targets that were not persued…but isn’t it all about what happens next?…they’ve got two platform generated molecules that will be in humans soon…lets see what happens.

        Btw, regarding 701, which blocks the PVRIG/PVRL2 interaction, has anybody wondered why there have not been any read outs from the phase 1 TIGIT monotherapy and TIGIT/PD-1 combo trials at BMY, MRK or Roche?…they’ve been in the clinic for almost two years, so they have the data.

        Maybe not blocking the PVRIG/PVRL2 pathway is why…and a triple combo may very well take brake off and step on the accelerator.

        Many folks in the field(Drew Pardoll, Elliot Sigal and others) think that it may very well work.

        Let’s hope so for all of the non responding and refractory PD-1 patients that they’re correct.

    3. Derek Lowe says:

      I will follow this with great interest, then. . .

      1. Jack says:

        That would be great…I look forward to the benefit of your insight…you always call them as you see them.


    4. Anonymous says:

      Do I have this right? Their AI pulled out the TARGETS (e.g., PVRIG, TIGIT, etc.) from among thousands of possible receptors among KNOWN pathways, as being important, targetable checkpoints? Once the receptor and its ligand are known, they optimize a mAB to inhibit binding. The mABs are the drugs. This is biology AI, not small molecule AI. It seems reasonable to me.

      They have done a lot work validating PVRIG, etc. as targets but I always worry about the missing colorful blobs and balloons (proteins; etc.) in the as yet unknown, potentially compensating, pathways, too. But I’m just a worrier that way. Hay, if it works, even just for a while, that’s still a good thing.

      1. Jack says:

        It’s drug development…if you’re not worried…something’s wrong!

        That said…their rationale is intriguing…notwithstanding Anon’s sanctimony.

    5. Design Monkey says:

      Upon checking both those advertised com701 and bay-whatever are antibodies.

      And would jackie be so kind an tell, exactly what (and how) supposedly AI there calculated? And how it was different from industry standart approach “let’s fuck whatever (semi) random XYZ thingy with a help of antibody and see what there pans out?”

      1. Anonymous says:

        I haven’t read any of their technical materials, just the cartoons and blurbs on their website. I am only guessing at this, but: I think the AI here is extensive BIO DATA analysis, not de novo small molecule modeling or protein molecular dynamics or enthalpomics 🙂 or anything like that. I think that they look at as much published (curated?) data as possible regarding experiments related to certain cell types or disease pathways: in vitro, in vivo, knockouts, knockins, mice, rats, alley cats, antagonists, agonists, stressors, different assays from different research groups, etc..

        Combining and sifting through all of that data, their AI identified PVRIG and some other targets as the most likely ones to be the most important in those pathways. “Ras shows up a lot, but it’s a red herring. Don’t waste your time on Ras. Focus on this PVRIG thing. It looks more important and targetable.” Having chosen a receptor and its ligand, design of a mAB becomes fairly routine but still needing optimization.

        I will make one more guess that I think is important. Whereas many programs focus on MUTATED targets (e.g., mutated “bad” Ras), I think that they are targeting normal PVRIG. In various disease conditions, many proteins are not genetically mutated and remain wt / normal, but their expression levels or co-localizations or post-translational modifications or other features are drastically altered. Thus, it IS possible to target a normal protein in the diseased cell because it is in a different environment than the wt protein in healthy cells.

        As I opined above, the downside is that some cells may realize that their wt PVRIG is being inhibited so they compensate by upregulating some other pathway. Every therapy has to deal with that possibility. (Gleevec was a wonder drug of 2001, but some patients become Gleevec resistant.) Even if you get a temporary benefit from one approach, that gives you time to switch therapies to another approach.

        1. Jack says:

          I think that is broadly correct…they discover proteins that their platform predicts to be an important checkpoint/pathway, then design high affinity antibodies to block them.

          And yes, cancer is masterful at using other pathways, which explains why the majority do not respond to PD-1 agents. Cgen’s belief is that by blocking both the PVRIG/PVRL2 and TIGIT/PVR interactions in addition to PD-1 blockade, DNAM is then free to upregulate the immune response.

          This is also the belief of Drew Pardoll and other KOL’s…so take that for whatever you think it may be worth.

  14. Isidore says:

    The co-founder is quoted as saying that she became “very frustrated with the drug discovery process” while enrolled in an MD/PhD program. I wonder, how many MD/PhD students have sufficient exposure to the drug discovery process to become frustrated (or enamored, for that matter) with it.

  15. luysii says:

    Just so you guys get an idea of just what “unmet medical need” really is, here is an excerpt of an Email from a friend and classmate (married 55+ years to his high school sweetheart)

    “XXXX is holding her own but on no meds for her memory loss at this time. She has her reasons and her psychiatrist understands and respects them although having pointed out clearly the implications of her decision to ‘hasten’ events along. It is heart breaking to me! “

  16. BernYeeBlackMockTurtleNeckShirtsLikeSteveJobsWore says:

    Don’t be so negative as they use one DRUG TO TARGET ALL GENES AT ONCE!

    I am so happy they found a therapeutic use for methacrylate, we can all strive to achieve the same health as those folks in BODY WORLDS.

    Ow, My Eyes!

  17. oldnuke says:

    Why is my Holmes-O-Meter pegged out? Is Henry Kissinger on their board by chance?

  18. SlimJimWins says:

    I understand why there is such skepticism over such bold claims from the founder. But have any of you ever tried to raise venture capital? If you do not make grandiose game-changing claims, you will not even get the attention of VCs. I have experienced this firsthand. It’s crap but those are the incentives VCs put in front of you. If your choice as CEO is to slowly starve or to pump up the hype, what do you think most entrepreneurs are going to do? So you can criticize Alice Zhang’s claims all you want (and they certainly do merit criticism), but at the end of the day you must understand that she is playing the game the way that the financiers want it to be played.

    1. John Wayne says:

      In my experience this sort of thinking is entirely correct. If your pitch deck isn’t full of 2-3 of the contemporary hot topics that will transform drug discovery nobody will give you a chance. Sometimes I think you have to … (don’t say, “lie’ …) … exaggerate … to get funding. Then, once you have the money you start believing your own BS and forget to do the thing you wanted to in the first place.

    2. Anon says:

      Maybe in the US, but in Europe, investors are more prudent/risk-averse, or at least averse to BS.

    3. Wavefunction says:

      I understand that you have to exaggerate a bit to get VCs’ attention and funding, but what happens when the rubber then hits the road? Are you going to then throw in the towel or pivot to something more realistic? If you’re doing the latter, at the very least then you’ll need to have some idea of what that more realistic thing is. I don’t mind having a slide or two that checks off the familiar buzzword boxes, but there needs to some modicum of real world recognition and challenges that will need you to retool that should be apparent at the beginning. Unfortunately there’s a huge amount of disposable funding in biotech these days, so VCs are probably happy to throw a couple of million dollars at high risk, high reward, fuzzily enumerated ideas. My problem is not really with the entrepreneurs, nor with the VCs – they’re simply feeding off each other’s expectations and playing the game – but with the rest of the world which seems to constantly fall for these pitches in spite of the usual red flags and tend to judge the worth of startups by funding and valuations. We’ve seen this movie before and didn’t like it, so next time it shouldn’t be unrealistic to start from a position of skepticism.

    4. Hap says:

      Are they doing their jobs though? If you fall for grandiose claims, claims that will likely have to be discarded when the company pivots to something else, then you end up not investing in what you’re telling your funders that you’re investing in, but ultimately in whatever the company pivots to. If the leadership is good and has good ideas (and the ability to work through them), then maybe that’s ok. For most startups, though, the leadership sin’t good enough to pivot to another good worthy idea, and so in investing in them, you’re going to at best mislead your investors and at worse lose their money. This doesn’t seem like a way to keep your job as a VC.

      1. SlimJimWins says:

        The entire business model of venture capital is tolerant to extremely high failure rates. If only a small fraction of their bets… errr, “investments” live up to the hype, they more than make up for all of the other failed investments. It is a very unique business where failure is the norm. The usual claim in VC investing is that 7 in 10 venture backed start-ups will fail, 2 will be breakeven, and 1 in 10 will be a home run. But that home run makes up for the other 9 “losers”. It’s just a numbers game for the VCs and individual flameouts are the expectation.

        1. Chester says:

          If only they could apply machine learning to lower that failure rate.

          1. Imaging guy says:

            Chester- that is funny. Maybe VCs could collect thousands of categorical and continuous predictor variables about the founders (age, sex, height, years of schooling, being a dropout or not, having a degree from Stanford, attractiveness, whole genome and RNA sequencing data of founders, plasma proteins profiles, and etc) and feed it into the computer. Dependent variable is successful start-ups (aka “exceptional responders”). The model (i.e. supervised machine learning) obtained can be used to make decisions so that 5 out 10 start-ups they give money become successful (“stratified VC model”). Maybe their model is so good that 10 out of 10 start-ups become successful (“personalized or precision VC model”). Later they can fund a start-up that will make a drug/drugs that will match potential founders’ protein expression profiles to those of successful founders (your plasma proteins A and B need to increased and C and D to be decreased). No capital would be wasted and the world would be a wonderful place full of successful peoples.

          2. SlimJimWins says:

            Imaging Guy- some funds are already applying machine learning to venture capital investing. These are two that I know off the top of my head, but I know there are at least a few others:


            Not sure if they’re up to RNA sequencing of founders yet.. That will be followed by scanning the brains of founders to ensure that their thoughts are sufficiently dedicated to “disruption”.

          3. Jonathan says:

            Funnily enough, Google uses an algorithm to do just that, but not one that uses ML by the sounds of it:

        2. Hap says:

          If the conversion rate for hyped hits is lower and the expected funding rate is higher, that leaves less money for bets that are either more predictably “blue-sky” investments (that aren’t as likely to pivot) or for the base load of less risky investments. In theory, you only need one to pay out big to be worth it, but you might not get there if your burn rate is too high.

          1. Imaging guy says:

            Jonathan- I have read your link. You might call their method machine learning (ML).
            “However, there is never a specific threshold wherein a model suddenly becomes “machine learning”; rather, all of these approaches exist along a continuum, determined by how many human assumptions are placed onto the algorithm.” (1)
            1) “Big Data and Machine Learning in Health Care”, Beam AL, 2018
            PMID: 29532063
            DOI: 10.1001/jama.2017.18391

  19. hn says:

    I’m fine with grandiose claims. But coming from a former grad student with no experience with drug discovery?

    1. Eka-silicon says:

      It *is* a bit rich isn’t it?!

    2. SlimJimWins says:

      This founder ticks a number of Silicon Valley boxes- young, photogenic, iconoclastic, went to Stanford, dropped out of Stanford, seeking to disrupt a stodgy old industry which doesn’t “get” that software is eating the world. If anything, the fact that she has no prior experience in drug discovery gives her MORE street cred in Silicon Valley- she is free of the baggage of said stodgy old industry.

      1. Gregr says:

        Yes, but dropped out of UCLA not Stanford

  20. HVL says:

    Meanwhile a Chinese company “quietly” announces positive phase 3 outcome for a natural product… It sure would be nice to see the actual data.
    Link is for the only news coverage I could find in English.

    1. Mister B. says:

      I found out the same press cover …

      Really surprised no one spotted an headline entitled “Algae cure AD” in a second-zone journal !
      Anyway, I hope they will have enough strong data to be analyzed seriously. If they have something new and potentially interesting, I wish they don’t ruin the trial.

  21. Bio bro says:

    The above comments are why reading biology journals is a bad idea. To much dogma….please read physics only, honestly.

    1. zero says:

      Plenty of dogma in physics. “One funeral at a time”…
      I think this happens when there is a substantial part of a field which is purely theory. The interface between theory and experimental result is where that dogma starts to get torn down, although humans are terribly resistant to rewriting their worldviews in the face of evidence.

      One thing I like about this blog is it tends to focus on results. When that’s not possible, articles are written with cautious optimism and a willingness to change opinions with new input.

  22. Arlen says:

    I could not resist commenting. Exceptionally wdll written!

  23. NolaGator says:

    There is a reason that you check the box for “seasoned management team” and “hundreds of years of industry experience” when you write a business plan.

  24. Anon says:

    Using functional or coding genetic data derived from patient samples to do target ID is not an outlandish idea, probably most big pharma are doing it. Coupling this with testing compounds in patient derived cellular assays e.g. iPSCs is also fairly mainstream. So behind the hype what they are trying to achieve is not unreasonable. Whether they can do it any better than many other startups in the same space eg BenevolentAI or eTherapeutics to name two U.K. based ones is a more pertinent question.

  25. anonymous says:

    Honestly, it is my perception that the “confluence of several trends” is partly consequential of the modern PhD educational process (or failure thereof) and academic culture. Let me briefly explain. If completing a postdoc is the new norm, simply because we produce too many PhDs, we imply to students, even after completing a rigorous PhD, that the scientific community does not find any value in their contributions/ideas yet. This creates a divide between enthusiastic young scientists and the experienced, knowledgeable scientists at the helm. There is a mutual lack of, dare I say, respect from either side and the current authorities on topics like drug discovery are resented and pushed against due to the perception that their advice is merely a cloak for hypocrisy, eliteness, and jealousy. We do not teach the value of men and women that came before us whilst also valuing a young scientist’s ambition or ideas. I digress. Also, let it be known that I believe a postdoc is a wonderful experience for growth in a number of ways. We simply have failed in the message the current culture is projecting.

    tl;dr: Young scientists with ambition and ideas, due to their environment, may feel like they have nowhere else to go but down this road.

    1. SearchMasterFlex says:

      Your statement presumes she was/is a scientist. Her style of pitching her so-called science could be used to peddle anything… she’s a CEO, she’s good at raising money, she’s photogenic, etc. etc. but I don’t see scientist anywhere in her credentials.

      However, this video may change your mind:

      1. Anonymous says:

        She says, “Drug discovery is still just a guessing game.” !!!!!! Not to worry, she is going to combine AI and genomics and wet lab — because no one else is doing it or has tried it ?????? — in order to transform the drug disco landscape. And she throws in personalized medicine, too; that should bring in another $10-20 MM. The key problem, Verge says, is choosing the right target, not making an effective drug.

        So THAT’S why so many drug candidates fail! WRONG TARGET! Doh!

  26. Gregr says:

    Yes, the truth is that Alice Zhang probably made a wise choice by dropping out of UCLA in terms of career satisfaction and potential lifelong earnings. Only 15% of PhD’s ever make it into Tenure Track jobs these days. The PhD’s who do make it into academics are actually starting to resemble entrepreneurs anyway. Many of successful faculty candidates already have funding lined up like R99 grants, and in fact that seems to be the new standard for getting TT positions at research universities.

  27. TMS says:

    This is an Andreesen-Horowitz slide deck on why “Software is eating Bio”

    I always chuckle when the consultants and VC guys describe the actual discovery process in this case carried out by “high salaried, unaligned lab drones.”

  28. idiotraptor says:

    From United States Patent Application, 20170360805
    Inventors: Chen; Jason; Zhang; Alice

    [0048] In some instances, isoflupredone and 15-deoxy-.DELTA.12,14-prostaglandin J.sub.2 are shown in a computational analysis to modulate genes associated with one or more of a motor-associated neurodegenerative disease. For example, based on a network pattern matching algorithm and the Connectivity Map database, isoflupredone and 15-deoxy-.DELTA.12,14-prostaglandin J.sub.2 are identified as modulators of one or more genes associated with a motor-associated neurodegenerative disease (e.g., see FIG. 1-FIG. 3). In addition, medrysone is identified to exert an opposite effect in comparison with isoflupredone by the network pattern matching algorithm (FIG. 1).

    If I understand correctly, the application is predicated on “computational analysis” showing that a synthetic steroid may “modulate” target genes implicated in a set of neurodegenerative diseases. Where is any supporting experimental data? I very quickly perused the published application and did not observe any (possibly I missed it). Mustn’t a successful patent exhibit in addition to novelty and utility, reduction to practice? Where is the latter? In the absence of credible supporting data, could this application be actually approved as a patent?

    Those who know IP/patent matters better than me can comment.

    Having just read John Carreyrou’s “Bad Blood”, I find the nascent Verge story eerily similar.

    1. Anonymous says:

      All of the claims are about small molecules and their use in the treatment of neurodegen diseases. The algorithm, software, etc. is not part of what is claimed. The figures appear to be from experimental data (gene expression maps, etc.) to support their claimed reasons for choosing specific compounds. (As if this approach is something new? Pfft. But they aren’t claiming the method of IDing drug candidates.) I’m not digging into details as yet (no lit access), but I’ll bet that this all cell based or, at best, mouse or rat based. This isn’t my field, but is that data from a grad student 6-week rotation project or is it more substantial than that?

      E.g., “[0038] FIG. 1 illustrates a summary of network gene expression changes induced by isoflupredone (left) and medrysone (right) in the Connectivity Map dataset. The intensity of node coloration corresponds to the statistical significance of the differential expression of that gene (green for down-regulated genes, red for up-regulated genes).” There are more of those and related Figures, 7 in all.

      The claims are, per usual, somewhat prophetic, referring to other small molecule derivatives or modifications that they probably haven’t actually made or tested yet. If someone else makes and patents a non-obvious derivative that actually works, I don’t think that Verge can stop them.

      I don’t have lit access at this time, but it looks like their lead compounds, isoflupredone and 15-Deoxy-Delta12,14-Prostaglandin J2 are kind of promiscuous signalling pathway and gene activators. “15-deoxy-PGJ2 (15d-PGJ2) is a metabolite of the PGJ2 prostanoid family that influences multiple signaling pathways by covalently binding with key signaling molecules. Among them, 15d-PGJ2 has displayed highest potency as an inducer of gene expression.” “The glucocorticoid activity of isoflupredone is approximately 10 times that of prednisolone, 50 times that of hydrocortisone, and 67 times that of cortisone as measured by liver glycogen deposition in rats.” and it is not approved for use in humans.

      From an endocrinology website, “The blood-brain barrier (BBB) is selectively permeable to steroid hormones: some of them (such as the gonadal steroids) can readily pass the BBB, while others (such as corticosteroids) pass only in small concentrations. This may however be modulated by hormone carrier proteins in the plasma.” However, if a drug has a proven beneficial effect, people will find a (patent protected) to deliver it.

      Based on the above, I suppose I should post something to Derek’s “Jadedness” topic.

      1. DonDraper says:

        People already get steroid injections for various things. Doesn’t delay any sort of degenerative process… after a few shots, often accelerate it.

        Amusingly it seems like Tim Draper, an original backer of Theranos, is also involved with Verge. Even when Theranos was clearly in the wrong, he saw Elizabeth Holms as the victim.

        Instead of “Bad Blood” maybe we’ll have a book about Tim Draper and his biotech failures titled “Oops AI did it again.”

  29. loupgarous says:

    How did DARPA wind up giving us work that actually led to self-driving cars? Modest cash prizes (compared to what university research cost us not to get results) and well-defined, rational, achievable goals.

    With all good will to Verge Genomics’ principals and investors, Grand Challenges aimed at achievement of discrete, well-defined, achievable technical goals have worked with investments much smaller than the average Big Research project. Once you transparently demonstrate that a concept works (which is where Elizabeth Holmes failed all along), THEN you start milking gigacapitalists for money.

  30. loupgarous says:

    Tangential observation – can’t find peroxynitrites mentioned under a story on neuroscience drug discovery! Not even once!

  31. Twice Shy says:

    I´ve nothing against a bolshy CEO and some good “to boldly go” marketing but what I am concerned about is the rights of the individuals from whom they are getting the longitudinal data and tissue samples. Usually these are given for “research purposes” but Verge have to make money with their findings so how are they remunerating the institutes that generated the data?

  32. Jack says:

    I would update my Cgen comment…both aforementioned drug candidates are now in the clinic; Identifier: NCT03667716

    These were discovered in silico by Compugen

  33. Yuva says:

    Thanks for this great discussion.

    I was doing competitor analysis for my company and was looking for more info about Verge.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.