Skip to main content

Chemical News

The Downside of Chemistry Automation

Automation in chemistry (especially industrial chemistry) is so pervasive that we hardly even notice it any more. (I have a whole talk that I give that’s partly on that very subject). But what is automation for? That’s the subject of this short piece in ACS Med. Chem. Letters by Jeffrey Pan of AbbVie. The answer would seem obvious, and fits into my rule of thumb that any question that can be phrased in the form “I wonder how come they. . .” has the answer “Money”. The obvious reason for R&D automation is indeed to save money, in those situations where repetitive tasks done by human beings can be sped up and standardized by machine.

But the easy opportunities for this sort of thing have been picked off by now. More capable equipment can take on more complicated jobs, but you have to decide if the throughput of the tasks it can do justifies its higher expense. We all depend on the systems that have shown their worth in labor and time savings (autosamplers!), to the point that we regard them as a necessary part of our work – but many readers will have also experienced the frustration of working with a machine that actually makes your life more difficult under the guise of helping you out. “Can you start this gizmo and then walk away and go to lunch without regretting it?” is a useful question to ask, and to answer honestly. And remember, “Sometimes” is equivalent to “No”.

The other choice is to make the machine’s task simpler and change your processes, but that doesn’t always work out. Pan notes the solid-phase combicham machines that were all over the place in the late 1990s and early 2000s, and how few of those techniques have survived. Solid-phase bead chemistry was a lot easier to automate than liquid-phase, but it also turned out, in the end, not to be as generally useful. The gathering of the dust on such equipment was a common sight as this realization set in. But the article makes some larger points:

. . .the science of chemistry is not advanced by having robots do tasks that would otherwise be done manually. Finally, innovation can potentially be decreased as chemists prioritize simple, automated work- flows over more complex, manual procedures. Encouraging chemists to perform more routine chemistry is unlikely to provide any competitive advantage.

I have to admit that I hadn’t looked at it from that angle, and I’m still turning this over in my mind. I’ve been thinking that as automation frees us up from routine tasks (with the definition of “routine” perhaps gradually changing over time) that chemists would be spending more time on the complicated stuff. If it can’t be done by machine, who else will do it? But Pan’s article has a different answer: if it can’t be done by machine, perhaps no one will do much of it at all.

His solution is to come up machines that actually make new modes of work possible, rather than just speeding up the old ones:

Rather than focusing on labor savings, we paradoxically explore strategies that encourage scientists to do more work. By developing systems that lower the hurdles to access new techniques, scientists are encouraged to investigate how and where to best apply cutting-edge technology. In that way, automation delivers a competitive advantage to our scientists by fostering chemistry innovation. We see automation as a tool to encourage the exploration of new methods. By reducing the cost of experimentation, we aim to lessen the reluctance to performing riskier experiments.

I certainly have no problem with that! As longtime readers will know, I’m generally a fan of new technologies and new synthetic techniques, and if they can be made more available and user-friendly in this way, so much the better. These can range from less-common forms of purification (benchtop supercritical fluid chromatography, counter-current chromatography, etc.) to more accessible flow chemistry for synthesis, better front ends for photochemistry, electrochemistry, etc. They try not to engineer anything from the ground up, but rather to enhance commercial equipment where some of that work has already been done.

There are still plenty of things to be ironed out, of course – one example given is when they set up apparatus to take advantage of a reported flow-chemistry diazomethane technique. Palladium catalysts will let you use that system for cyclopropanation reactions, but the AbbVie team found that the gas-permeable tubing needed for the reaction conditions tended to foul in the presence of the Pd reagents after multiple runs. They reworked it successfully to where the catalyst comes in after that part of the reactor, but this is the sort of thing that you’re only going to find out with experience.

There are numerous other examples given in the paper, with references, and I encourage any fan of automated chemistry to go over them and learn from the AbbVie work. There are similar efforts going on at some of the other large companies too, of course, although not all of them have the explicit “automate things that people aren’t doing at all” mandate. That brings me back around, though, to the question above: does the automation of routine chemistry lead to people spending more time on just that routine chemistry, or does it free up other work?

Comments are welcome. My first thoughts are that such automation may well encourage “the same stuff, just lots more of it” when that is sufficient to solve the problem at hand. If a med-chem project has been needing a way to crank out Buchwald-Hartwig couplings to explore a key SAR feature, than a machine that lets that happen more efficiently is good news. But a machine that encourages everyone to do mostly Buchwald-Hartwig reactions to the exclusion of other chemistries is probably not such good news, overall. That’s what happened with the first generation of Pd bond formation; the compound lists filled up with aryl-aryl couplings, and that’s not the answer to everything (far from it).

But the tricky part is that a lot of medicinal chemistry problems are solvable with enough brute force. Often enough, there is a particular analog that manages to put together properties that are sufficient to advance a project – not perfectly, perhaps, but good enough. I think the important thing is to strike a balance, to enable enough different kinds of chemistry so that same brute force can be applied in many different ways to produce a wide variety of structures. That will both broaden the chemical landscape and reduce the temptation to overuse some particular method just because it has a bigger motorized crank attached to it.



32 comments on “The Downside of Chemistry Automation”

  1. Nameless says:

    But is the newest gizmo cheaper than 3 chinese graduates, desperate for a job?

    1. ScientistSailor says:

      I made that point to some RevMed people a while back. I have a machine that works overnight for me in China, AND it can unclog it’self…

  2. Barry says:

    One niche in which automation has been used very well is NMR. There are real time savings in not having to walk my own sample down to the NMR lab, and carefully lower the sample pneumatically into the magnet and record the spectrum. But I remain free to get all the not-quite standard answers (difference double irradiation, NOESY…)too.
    The trap is that a med. chemist may feel the pressure to submit the hundred analogues that can be cranked out by semi-automated or automated Suzuki couplings rather than the one analogue that requires chemistry that doesn’t fit the machine.

  3. Anonymous says:

    Don’t tell Lilly. Automation has been one of the recent excuses for getting rid of AMRI (and Lilly) chemistry resources.

    1. Anonymous says:

      Exactly. Excuses for, not reasons for.

  4. CombiCham says:

    ”innovation can potentially be decreased as chemists prioritize simple, automated work- flows over more complex, manual procedures”. This exact behavior has been seen the last three decades…and it doesn’t have to be bad, overall. Chemists may be conservative but they sure aint stupid.

  5. Kelvin says:

    I’ve always had an issue with the term “innovation process”, which seems like a contradiction in terms since a process is nothing more than a specific sequence of constraints, whereas innovation should be about breaking free from any constraints. Thus we need to innovate the way we innovate by exploring alternate processes.

    Therefore the key for innovation is not to automate a given process, but to automate the exploration of *different* processes.

    1. John Wayne says:

      Kelvin, you make an interesting point; I think I agree. This may also highlight the inherent tension between automation and creativity. Computers and robots are really good at mercilessly doing the same thing, while creativity often wants you to do something different. The interface between the two is hard to find, but possible.

      So, what are the things that have become automated that are great?
      1. NMR robots and software
      2. Iterative injections and fraction collection on chromatography and purification instruments
      3. Automatic pumps for your rotovap
      4. Some synthesizers once you have a system worked out.

      Seems like the common pattern here is brute force – what is where computers and robots are at their best. The limitation is, you have the have the system fairly worked out before you can brute force it. So … if you are planning to do some cutting edge science at your company any suggestion that automation will save the day should be viewed with some suspicion. I like to call it, ‘guilty until proven innocent.’ Some of it is good, but not most of it.

      1. Kelvin says:

        Perhaps just like evolution by randomization and recombination of genetic code, computer codes and algorithms should also be subject to some (controlled) degree of randomization and recombination? 🙂

  6. zero says:

    Automation is often sold as a way to reduce labor requirements, not as a way to reapply an existing labor pool. The first point of view is the one most likely to succeed for a sales rep or a manager on a budget crunch, which is why that outcome is common. The second point of view is still possible, but you have to go into it expecting to spend money on all your existing labor plus more money on the new automation.

    This is not unique to pharma; the same dynamic exists in pretty much anything software-related.

  7. anon the II says:

    I was heartened to read Jeff’s opinion on the overly integrated systems as described in reference 4. Expressing an opinion similar to Jeff’s to one of the authors of that paper probably costs me my career in big pharma. I think I asked him if he’d actually ever done any organic synthesis.

    Lessons learned.

  8. Scott says:

    ” “Can you start this gizmo and then walk away and go to lunch without regretting it?” is a useful question to ask, and to answer honestly. And remember, “Sometimes” is [*always*] equivalent to “No”.”
    Fixed that for you.

    “if it can’t be done by machine, perhaps no one will do much of it at all.”
    And that is a scary thought.

    For those talking about their automatic overnight, self-unclogging ‘machines’ located in China, how good is their quality and repeatability? If it’s really good, on par with the machine someone is trying to sell you, then you’re probably not in the market for a machine to take up space in your lab. But if it’s not the equal of the machine, I’d seriously consider buying a machine and a post-doc (or even a BSc) to run it!

    1. achemist says:

      thats not necessarily true.
      e.g. Peptide or oligo synthetizers, they run perfectly fine without someone there (and they better do given the synthesis of long peptides takes smth like 2 days).

    2. outsourcingboondoggle says:

      Does your machine tell ridiculous lies to your face like you’re some kind of an idiot?

      1. John Wayne says:

        LOL! It has been a few months since that has happened to me. I’m overdue.

        But on a serious note, instruments lie to nonexperts all the time. I literally cannot count the number of times I’ve been informed about a result from LCMS data only with no considerations of extinction coefficients and/or probabilities of MS detection. Whenever anything is a black box to a user, they will believe the occasional spectacular lie. I’m not sure if the machines have enough emergent intelligence to lie more to less skilled users, but people certainly do.

        1. msjunkie says:

          with regards to erroneous MS data 3 words…
          extracted ion chromatogram

  9. beentheredonethat says:

    People keep flogging this dead horse, the only place automation is applicable to discovery chemistry is in purification, unless of course you think there is merit to generating reams of amides or other simple chemistry but even then the variability of the solubility of reagents and products makes is not worth it.

    1. Scott says:

      I disagree.

      There are times automation can be useful, usually in the brute-force attack on making something steps.

      I’m not a chemist, but if you’re doing something that can be set up into a 100-well sample tray and you’re looking for differences or something similar, automating the differences can be a real time-saver in terms of your overall labor levels. The labor that required a human to do was to set up the steps, the robot can do the process.

      1. painter says:

        you said it yourself….you are not a chemist

        There is very little applicability of automation to the kind of discovery chemistry that goes on in a med chem lab.
        The picture on the front of the aldrich catalog every year is a painting by one of the masters.
        could you “automate” the creation of a fine painting?

        1. Design Monkey says:

          Significant part of work of fine painters of the past was making portraits of various highborns and other fat cats of society. That automatizes just fine with photography and even makes portraits available to everybody.

  10. Hap says:

    Maybe the question should be (as it seems with lots of pharma R+D) that if you don’t really care about the quality of what you’re doing, why are you doing it?

    Making something that allows you to do lots quickly but doesn’t allow to do the things that matter to research (while getting rid of the people who could) suggests that the people running pharma don’t assume that the quality of output of research has anything to do to with what their companies’ outputs are. In that case, why spend the money at all (other than to create a Potemkin village to justify their salaries, which I’m sure warms the hearts of investors and patients everywhere), If that’s not the case, then management needs to stop confusing activity with achievement.

    1. shakingmyhead says:

      LOL, so you mean to tell me that my supervisor heaping accolades on his cheap compliant labor for generating large numbers of useless compounds is wrong?

      1. Hap says:

        That’s what he’s getting paid for, though. I would like to hope that he’d stop getting paid for that, but since his bosses are, and probably his bosses’ bosses, too, that’s not likely.

  11. tom says:

    To paraphrase Mitchell and Webb, “Chemistry is in dire need of mechanisation”.

    One of my most disheartening moments (which led me to eventually leave the field) was at pacifichem 2015, specifically a Baran lecture. It wasn’t necessarily the ‘Human Wave/40 hours in the lab straight’ kinda stuff that they were pulling, but the fact that it was actually effective in getting things done.

    In Maths, the limiting factor on progress is your own stupidity. In Chemistry your ability to come up with things to try will always outpace your ability to try them by an order of magnitude at least.

    I have long fantasised about a system, similar to how computational chemistry on a cluster is often done, where you plug in the parameters for ten or so reactions in the evening, let the robot run them overnight, and wake up and analyse the resulting spectra in the morning and plan your next ten or so reactions. Measurements would be precise, repeatable and automatically entered into a database. Software (and occasional hardware) updates would allow the system to cope with new and tricky reactions, and machine learning could automatically be applied to reaction results, updating on the fly and providing a best guess for what to try next to optimise the reaction. Benchwork would be relegated to the most unusual and difficult of reactions.

    1. PM at NCATS says:

      Exactly this. Sometimes brute force is impossible and impractical. The idea of having standardized, reproducible chemistry sounds extremely appealing.

      Interested in trying to make this dream a reality? Have ideas for how to combine ML/automation/HTS to create the lab of the future? Check out NCATS’ ASPIRE program: Currently running Challenges with $2.5 M at stake!

  12. Absolutes zero says:

    My recollection is that when Gary Kasparov was beaten by IBM Blue, he came to the conclusion its not just machine or human- innovation comes from combining both! I tend to agree with the Chess master. It’s all about how we work together to channel chemists to innovative chemistry.

    1. Design Monkey says:

      Well, Garry had his chess behind kicked a long time ago. AI marches on, and nowadays Alpha Zero will beat him, completely without any chess master teaching it.

  13. Daniel Yohannes says:

    When I was a graduate student with Dale Boger, I had the unique privilege of doing a fair amount of novel high-pressure chemistry (at multiple kilobar pressure; I used to set up my reactions with Queen/Freddie Mercury singing “Under Pressure” as backdrop music). This was a way of activation of chemical reactions through a means not involving thermal sources, which made certain chemistries possible that were not possible by thermal activation. The downsides were that the machines that made this possible were expensive, uncommon, a bit messy, and not amenable to scaleup. But man could this technique access chemistry that was otherwise hard to access. I’d think this would be an excellent test-bed for “automate things that people aren’t doing at all (or hardly doing at all)” mandate you speak of, no? Automate and Scale, that is.

    1. Russ says:

      Diamond anvil?

    2. nooneislisteningtoyou says:

      LOL this is coming from one of the arqule clowns

      They would coach the chemists to lie to customers about doing chemistry with “robots” when it was painstakingly done by hand

    3. cynical1 says:

      So, it was the Queen music that provided the activation energy for the reaction?

  14. M. Striker says:

    So, my vision of chemist-supervised robotic armatures doing basic chemistry at a hundred workstations is merely a pipedream?

Comments are closed.