Skip to main content

Clinical Trials

What’s Crucial And What Isn’t

One of the reasons that people in or near this business can write such gaudy press releases is that it has so many moving parts. That lets everyone claim that the part that they’re addressing is Crucial. Think of a car: the wheels are indeed key to mobility, but so is the engine. As is the oil, the power source (be it gas tank or battery), and any number of other parts. Or you can use the human body as a metaphor: there are parts of it that you can cut off or out, but there are quite a few parts of it that you can’t. All of the latter have a plausible claim to being crucial.

Let’s consider, though, this claim which is the headline of a Stat story from the last couple of days: “Amazon, Google, and Facebook are using AI to find protein structures”, it says, “The tech giants are becoming increasingly active in deciphering protein structures, a crucial step in helping to find drugs” We heard a lot about how crucial this was during the last protein-folding competition as well. But is this true?

I’m going to sound like a heretic (at least to the people doing that work!) and say no, not generally. There are cases where such structures can be very helpful, but there are cases where they don’t do you that much good at all. I would not want to do fragment-based drug discovery without access to ligand-bound structures, for sure, although it’s certainly possible if you don’t mind wasting a lot of time feeling around in the dark. But if you have a solid primary assay against a target you really believe in, and an animal model with real translatability, then no, you can push right ahead. Don’t you need a defined target to get a drug to market? No, you don’t, actually, although it certainly can help. What you absolutely need are safety and efficacy data. And those you need successful human clinical trials, and to get to those you need for the FDA to allow your IND application. The best way to get one of those approved is to show activity in a relevant animal model and clean toxicology studies in at least two species, and you can do those without knowing the protein target at all, let alone its structure. There aren’t enough reliable animal models in the world for this to be a common route, of course, but it’s certainly possible (and many years ago, before my time, it used to be the rule).

Even as you read through the Stat article, this point becomes apparent (to their credit). Protein structure determination simply isn’t a rate-limiting step in drug discovery in general. If we were (far) better at modeling potential drug candidates to such structures, that would make things a bit different, but we’re really not at that point yet (cue the AI people who are doing such virtual screening!) Both of these fields – prediction of protein structure and prediction of small-molecule binding – are advancing, but neither of them are ready to be the backbone of a company’s drug discovery efforts yet. They’re tools, and they’re sometimes very useful and sometimes a waste of time and effort, which you can say about most of the other tools as well.

I welcome the big-tech folks to the protein structure party, of course. We really do need more insights in that area, and if such predictions are every going to be generally useful we’re going to need all of the insights we can get (and all the processing power we can get, too, most likely!) But if someone is telling you that protein structure prediction is going to lead to a big leap in drug discovery efficiency, hold on to your wallet. What would lead to such a leap?  Off the top of my head, I’d say better prediction of useful drug targets, more translatable disease-predictive cell and animal models, and earlier assays that are more predictive of human toxicology. Those, as far as I’m concerned, address the real killers in the whole process. Protein structure just isn’t on that list.

33 comments on “What’s Crucial And What Isn’t”

  1. GrammarCheck9000 says:

    “Or you can use the human body as a metaphor: there are parts of it that you can cut off or out, but there are quite a few parts of it that you can’t. All of the former have a plausible claim to being crucial.” – wouldn’t that be, “All of the latter…”?

    1. Derek Lowe says:

      It sure would! I transposed sentences and parts of sentences in that paragraph several times, and that one never got flipped back. Thanks!

  2. comment_is_free says:

    I agree this is not the rate-limiting step in drug discovery, and the article is exaggerating a little there. But at the same time I think we have to agree it’s very cool research that if successful COULD down the line change the way we do drug discovery. Advances that solve the problems we care about don’t always come from where we expect. It’s interesting to me that of all the problems in biology, THIS is the problem the tech companies have decided to tackle too…maybe it’s just the one most amenable to machine learning at this point

    1. zero says:

      It’s possible that this work is a critical step for the Silicon Valley crowd on the path to improving the rate-limiting steps Derek pointed out.
      They need to pass med school before they can manage a residency. We’re hoping some day they get their MD and open a practice, but that’s a long ways off right now.

    2. annon says:

      Almost everything in machine intelligence needs to be standardized. No phenomenon in a given Biological experiment follow a standard procedure.

  3. Carlos Montanari says:

    Does this really work in practice? We just got a project rejected by the NIH for exactly that reason. How many drugs are in therapy for which we do not know the mechanism of action?

      1. JK says:

        If you change that to “how many drugs are in therapy for which do not know the complete mechanism of action?”, I would say the answer is pretty much all of them.

        1. John Wayne says:


        2. cynical1 says:

          Disagree. I’d say that we have a pretty good handle on how most anti-invectives work.

          1. cynical1 says:

            Sorry about the auto correct: anti infectives.

          2. Curt F. says:

            Anti-invectives could represent a major unmet medical need though.

          3. Andrew Molitor says:

            Fuck you.

          4. Dionysius Rex says:

            Well played there Mr Molitor!

          5. cynical1 says:

            @ Curt F. I’m not suggesting that there is not unmet medical need in the anti-infective area (e.g. TB) or that you could not develop a drug with an unknown MOA as an anti-infective. I’m just saying that I think that the scientific community understands how most anti-infectives on the market work, that’s all. For instance, an HIV protease inhibitor inhibits that enzyme and inhibition of that enzyme represents the totality of it’s efficacy as an anti-viral against HIV.

      2. Mol biologist says:

        IMO Felix Meerson was outstanding talented physiologist to ask questions and to find a connection from particular to general.
        IMO Felix Meerson was an outstanding talented physiologist to ask questions and to find a connection from particular to the general.
        It is particular that cancer cells are not “fighting back”, in general, they have that behavior due to the absence of resources to do anything…It is particularly the mechanism of PHD hydroxylation and HIF stability (which is correct and congratulation to all 2019 Nobel Prize winners).
        However, in general, I would be referring to the earth evolution process and imply again that PHD can’t feel the oxygen level due to other ancient functions.

    1. JB says:

      Many drugs in the pipeline have undetermined MOAs. I get that this blog is mostly focused on small molecules and chemsitry, but when dealing with biologics like cell based therapies, you aren’t required at all to understand mechanism. There are tons of companies that exist right now that are trying things like delivery of bacteria to the gut with the hopes of producing cancer benefits. Somehow there is synergy with the immune system when you deliver bacteria to the gut that will help with cancer. There are also hundreds of clinical trials with stem cells. The explanation is always that stem cells like MSCs secrete magical cocktails of factors that they secrete that produce benefits for almost every indication under the sun. Target based modulation with a defined MOA is something mostly small molecule people do, but more complex biologics are often the wild west.

    2. Anonymous says:

      There are examples of drugs that were thought to operate by one MOA but then shown to actually operate by a different MOA … until future experiments shed more light on the complexity of the situation. Not surprisingly, the drugs still work, even by the “wrong” MOA! Thankfully, the drugs didn’t read the papers.

      Methotrexate was developed in the late 1940s to inhibit folic acid synthesis (for the treatment of cancer). It IS an inhibitor of dihydrofolate reductase. It was not developed from a protein structure. It was developed as an analog of the natural substrate. It was a miracle drug in the 1950s and continues to be a very good treatment for some cancers (and other conditions by DIFFERENT MOAs). However, when they finally got an X-ray of the methotrexate-DHFR complex, it wasn’t not binding like the natural substrate at all! Those idiots designed a successful drug from a proven to be incorrect hypothesis.

      In the late 1980s-1990s, pregagalin was developed as a GABA analog in order to have an effect on GABA receptors. In the early going, it produced outcomes fully consistent with that MOA. Brilliant thinking! But it turned out later that it has almost no effect on GABA or GABA pathways. It is an ion channel blocker. Doh! As far as I know, nobody returned the royalty checks. 🙂 There is probably more to learned about pregabalin and further details of the MOA.

      Tack Kuntz’s DOCK program successfully predicted structures of strong binders for a target protein. They were synthesized and tested and were tight binders. Then, they got the X-ray structures of the complexes and found the compounds were bound in a completely different location and bound in different conformations than they had predicted. (Laymen: Picture the small molecule cursive E sticking to the pointy bits at 11-12 on a clock. It turns out the curly parts of the E were actually sticking to the curvy parts at the 5-6 on the clock.)

      Protein structures were unnecessary for methotrexate, pregabalin, the Kuntz example and others. However, compound-protein complex structures were helpful in figuring things out.

      I prefer “currently accepted MOA” or “consistent with” or “supportive of” a particular MOA.

      What is crucial? Honest scientific researchers that perform experiments properly and reproducibly, have respect for and follow the data, question and retest suspicious outcomes, and … I’ll just refer you, once again, to Feynmann’s essay on Cargo Cult Science.

  4. Torvis says:

    I’d think that target selection has a lot to do with structural determination not being the rate-limiting step–surely most or all of the targets that people choose for target-based drug discovery have been previously crystallized and iterating that structure with more new ligands is relatively straightforward. There are lots of difficult-to-crystallize targets out there for which structural determination would be the rate-limiting step, it’s just that no one is doing target-based discovery on them for that reason.

    Now, I’m not saying structure predictions are likely to be particularly helpful in this regard, but I think it’s a point worth thinking about, especially since you bring up “better prediction of useful drug targets” as an example of something that would be useful. I’m sure there are a lot of useful drug targets out there about which it is challenging to get good structural information. Would structural predictions be useful for those? Probably not very, especially if they aren’t based on any existing structures, but as someone who works on such a protein, having a model seems better than having none at all.

  5. Jb says:

    And then there is the whole problem with the fact that proteins are post-translationally modified. Yeah, it’s cool you used AI to predict protein folding, but the shape of the pocket changes when hundreds of PTM combinations on proteins significantly restrict conformation populations, increase/decrease solubility, and have a slew of other physiochemical effects. Heck, there may even be large PTMs sitting directly in the pocket of your protein, or the half-life of a protein target is largely regulated by a ptm.

  6. Curt F. says:

    A common theme here at ItP and in “establishment” medchem folks writing elsewhere is that some exciting (or maybe just overhyped) new technology is *not* rate limiting for drug discovery. This includes protein structure determination, fancy AI approaches to novel structure generation, etc.

    We hear a lot about what is *not* rate-limiting, but far less about what *is* rate limiting. Is there a coherent summary of what is rate-limiting in drug discovery? I think I remember asides and snippets from past posts that suggest maybe biological understanding is what is rate-limiting. Is that right? If so, can that be elaborated?

    I would very much appreciate pointers from Derek or other cognoscenti to thorough discussions of what is rate-limiting in drug discovery. Maybe there is already a post on this from years back that I missed?

    1. Frank says:

      Here you go.
      Rate limiting steps
      Human trials (1~6 years), CMC ( at least 1 year), toxicology in NHP (6 months ~ 1 year), animal models (1 year if designed in parallel).
      Screening for molecules /antibody campaign ~ 6 month
      Regulatory filing and FDA review for approval (~1 year, but you are happy at this point)

      On the other hand, solve a crystal structure by WET experiment (if successful, less than 1 month, and very cheap!)

      The problem of drug discovery these days are
      1) Lack of “real” targets that are human-relevant, but not the lack of structure
      2) Lack of “real” animal models that predicts, as Derek points out
      3) Lack of “speedy” enrollments, as clinical trial enrollment rate is fairly low (even in the US)

      1. Curt F. says:

        Thank you for your thoughts, Frank. Much appreciated. It seems like 1 and 2 are areas where academic research could contribute mightily, especially 1. It also seems like there is no tried-and-true approach to improving either of these, but instead that the process is a messy one of proposing new targets and disease models in e.g. the literature, and then waiting a while to see if you were right. (That’s a long winded way of saying “science”, I guess.) If other folks agree I look forward to many new posts here and elsewhere on exciting new disease models and candidate targets.

        1. johnnyboy says:

          If by rate-limiting you mean what causes drug programs to crash and burn, I think it’s fairly well-established that it’s human trials where most drugs go to die, particularly phase 1 or early 2. That can be from lack of PD response, poor PK, unexpected/unmanageable tox, or lack of any hint of efficacy. Of course since companies invest heavily into clin trials, there has to be solid pre-clinical evidence to support going into human – so failure at the clinical phase essentially means that the preclinical methods you’ve used to support going into human failed somewhere. So as Derek pointed out above, poor predictivity of preclinical assays (for efficacy, tox, or PK) is the problem. There is no quick fix for any of that, apart from doing better science. And doing good science can be at odds with corporate management that’s fixated on attaining pre-defined (and arbitrary) goals/scorecards/targets for the year. If you’re in a team that’s been designated to get through a certain milestone by a certain time, you’re not always inclined to give a receptive ear to a member of your team pointing out that some of the data used to get to that milestone is crap.

          1. Invisible man says:


            So, perhaps an open discussion on what you call “better science” might help.
            IMHO, to uncover novel, translationally relevant and accurate insights we need to start asking different questions. These types of initiatives seem like trying to answer a question for which we already know the answer in good enough detail or is not even necessary to move a project forward. Clearly not crucial.

            We are looking under the lamppost and there ain’t no keys there… at least, the keys that will open the door we aim to open…

  7. MoMo says:

    Curt F.

    What’s rate limiting? More Molecules!

    Been saying this for years on this blog and as usual in human nature, any means to cut corners, feed new fads, and basically find new ways to skirt REAL WORK is a constant here in this industry.

    So go ahead, mine data, suck up to AI and other vowels, but the real innovations happen by those that work in the lab, creating new molecules- and without looking at the clock or being influenced by trends that make the weak feel empowered.

    MOre MOlecules

    1. Curt F. says:

      Thanks for the reply! Would it be fair to interpret your response as saying library size is what is limiting then? As in, we have enough good high-throughput screens, but not enough molecules to put in them? The need is just for more structural diversity?

      1. MTK says:

        Curt F.,

        That’s not how I interpret MoMo’s statement of “more molecules!” It’s not necessarily we don’t have enough in libraries but rather we don’t have enough after HTS. So it’s not a lack of starting points, but not enough molecules made and examined after the appropriate starting point is identified. Of course, each of these can then be fed into screening libraries also.

        MoMo, you can correct me if I’m wrong.

  8. Tc says:

    Is this article really justified when >80% of drugs designed largely by the methods championed fail in clinical trials? Pretty pompous for people with such a dismal record.

    1. Ursa Major says:

      This post is saying that you don’t need to know the protein structure to bring a drug to market, not that it doesn’t help. Describing it as ‘crucial’ is an exaggeration.

    2. loupgarous says:

      @Tc – the article reflects actual experience bringing drugs to market (>90% fail somewhere between animal tox/PK and human clinical testing). Whose word will you take as to the “crucialness” of knowing protein structure but that of people who’ve spent their lives making drug molecules and watching them progress through the evaluation process. It’s not pomposity to state a fact – awareness of protein structure doesn’t really impact success or failure of a new investigation drug in the vast majority of cases.

  9. Mol biologist says:

    I am wondering how modern landscape of drug discovery especially for cardiovascular market in US would look these days if Felix Z. Meerson get Nobel prize for the concept of adaptive stress. Unfortunately he died in 2010 at Kanzas nursing home……

Comments are closed.