So I go away for a few days, and people are already planning to replace me (and my colleagues) with robots? Can’t turn your back on anyone in this business. What that article is talking about is the long-term dream of a “synthesis machine”, a device that would take whatever structure you fed into it and start in trying to make it. No such device exists – nothing even remotely close to it exists – but there’s nothing impossible about it.
A British project called Dial-a-Molecule is laying the groundwork. Led by Whitby, the £700,000 (US$1.2-million) project began in 2010 and currently runs until May 2015. So far, it has mostly focused on working out what components the machine would need, and building a collaboration of more than 450 researchers and 60 companies to help work on the idea. The hope, says Whitby, is that this launchpad will help team members to attract the long-term support they need to achieve the vision. . .
. . .Some reckon it would take decades to develop an automated chemist as adept as a human — but a less capable, although still useful, device could be a lot closer. “With adequate funding, five years and we’re done,” says Bartosz Grzybowski, a chemist at Northwestern University in Evanston, Illinois, who has ambitious plans for a synthesis machine of his own.
There’s a lot of room in between “as adept as a human” and “still useful”, let me tell you, and that word “useful” covers a lot of ground all by itself. But I agree that there’s a lot of potential here, and that we’re at least getting close to the point of being able to realize some of it. But then the arguing starts. This topic came up here just recently, and has before as well. But those examples were far more simple than the Turing-machine like ideal synthesis device.
That one really is decades away, I’d say. There are a lot of problems to be dealt with. For one, you have the physical handling of reagents and reaction purification. Steve Ley’s group and Tim Jamison’s group are two of the best at this, and I recently had a chance to hear Jamison talk about some DARPA-funded work he’s been doing on an automated platform to make simple pharmaceuticals on demand. It has taken them no small amount of work just to get the dispensing, purification, and transfer parts to mesh correctly, and that’s for compounds for which defined synthetic routes have been well worked out. Throw in the problem of figuring out a good synthetic plan (among the vast number of reaction and reagent possibilities) and the problem gets wildly, exponentially more hairy.
That last link discusses some very interesting work from the Grzybowski group, who have been developing a system called Chematica. Here’s the latest on that:
These demonstrations have impressed synthetic chemists, although few have had a chance to test Chematica. That is because Grzybowski is hoping to commercialize the system: he is negotiating with Elsevier to incorporate the program into Reaxys, and is working with the pharmaceutical industry to test Chematica’s synthesis suggestions for biologically active, naturally occurring molecules. Grzybowski is also bidding for a grant from the Polish government, worth up to 7 million złoty (US$2.3 million), to use Chematica as the brain of a synthesis machine that can prove itself by automatically planning and executing syntheses of at least three important drug molecules.
I’ll watch this with great interest, but it’s worth noting that no one has yet tried to hook Grzybowski-type software with Jamison/Ley-type hardware. And that’ll be a real joy to execute. This looks to an outsider (me) like one of those cases where the software folks figure that the hardware is pretty much ready to go, and the hardware folks might be figuring that the software is more or less at the late-debugging stage. I suspect that anyone thinking down those lines (and there are some quotes in that Nature article that suggest it) is in for a rude shock. The kinds of reactions that useful software would suggest will be things that no one has ever tried to automate, and the range of reagents that can be accommodated by the existing hardware may well cripple the software algorithms before they even get off the floor. There’s a lot of work to be done.
My guess is that we’re going to see many years of machines that can do some things in very well-defined areas, but which will prove useless (or worse) if you try to push them into unknown territory. And unknown territory is what it’s all about. The article mentions the most difficult level such a machine could work at: synthetic routes where new reactions have to be invented. Don’t hold your breath for that one: a machine that could work its way through the first semester of an undergraduate organic lab course would, by current standards, be a tremendous accomplishment.
But at the same time, I want to re-emphasize that there’s nothing intrinsically impossible about any of this. It’s just crazily hard, and will require years of machete-hacking through thickets of engineering difficulties. I think that this really is the direction the field is heading, but it’s going to be a long, long road.
Among us human automatons, is there anyone that uses or used LHASA (Corey), SYNGEN (Hendrickson), Synchem (Gelertner), or others? I think SKF licensed Synchem in-house for a few years.
When are they going to automate chem-blogging and chem-commenting? Then I’ll really have nothing to do.
Are robo-chemists another entry in your “Things I Won’t Work With” files?
Will the robot chemists be cheaper than the Chinese ones?
So the cost of not-finding important new drugs will be reduced. Good-oh!
Not sure if I am missing something here but according to this quote:
“I would consider it entirely feasible to build a synthesis machine which could make any one of a billion defined small molecules on demand,” declares Richard Whitby, a chemist at the University of Southampton, UK”,
they are not promising to make a molecule by demand but rather to choose one from the defined combichem catalogue. Anyone interested in the project still?
My favorite was William Jorgensen’s CAMEO which worked off of first principles rather than a huge database of reaction conditions. It was very interesting and also could propose new products given starting materials and reaction conditions. It was sometime useful in elucidating what might have gone wrong in a reaction. The CAMEO program was really a nice AI type system.
Unlike chemists where, as Derek notes, the magic machine is pretty much blue sky at this point, neurologists in the 70s were faced with a machine that could find things earlier than our clinical examinations. Much fear from all concerned. However, we’re still in business.
Ditto for telephone operators. The communication system probably employs more people now than it ever did.
For details see https://luysii.wordpress.com/2014/08/07/as-if-the-job-shortage-for-organicmedicinal-chemists-wasnt-bad-enough/
For everyone saying that this thing is no good, I think that it certainly has its limitations (like it probably wouldn’t be very good at making some huge natural product with tons of stereocenters and stuff, but those algorithms certainly could give you worthwhile ideas that you wouldn’t have thought of), it definitely would have a domain of use. For example, it could be used to make starting materials or reagents that are of low to moderate complexity but aren’t commercially available, or are expensive, giving you more time to do other things. The project I was just on was a pain in the ass because we had to make a lot of our substrates. It wasn’t particularly difficult, but it made everything take about 1.5x as long. I also can see this as greatly improving the situation for labs that don’t have heavy synthetic knowledge to know how to do a lot of these things themselves.
One of the biggest things I can see this being good for is non-organic groups (like physical or inorganic) that want to do really intense isotope labeling studies, like say you want to make a commerically available ligand with selective C13 and deuterium labeling? That could require some pretty heavy organic knowledge and skills to know how to do.
And if you encounter something that the machine couldn’t do or couldn’t get to work (which would probably be pretty common)? Well time to go to the old fashioned way.
As long as I’m not the one who will have to clean/fix it.
#3, What is the difference?
Flash forward to 2060, when Shanghainese rock bands will be singing “The Ballad of Jian Qiang”, who died with a pipet in his hand, lawd, lawd.
Perhaps If you could build one the size of a vending machine, that could do simple compounds, you could install it in any campus town. The premeds I went to school with would all order up product to turn in rather than do all the menial labor themselves. They plan to pass off all the menial labor to their nurses later, so this could be like on-the-job training.
But seriously, I could envision scenarios wherein a hazardous chemical could be synthesized on-site. I would not want to calculate the ROI for that purpose, with the difficulties Derek outlines. Great post, BTW
How about a cook machine that allows me to dial up a Kung Pao Chicken, Prof? Isn’t that more useful than the fancy schmancy Maitotoxin?
You’ll know the robo-chemists are getting close when they all start writing in the comments to your blog swearing at all the unreproducible syntheses in the literature, and the endemic yield inflation which messes up their analyses. Or wait…maybe they have already?
Or are the authors, say Grzybowski, going through the literature, running each reaction, and noting whether they can reproduce it and what yields they think they can get, and what the real price of reagents is after you eliminate vendors that will sell you crap? (I’m guessing he’s still in the process of talking to others to find what others have reproduced, which works for what? maybe 40% of the literature? and the rest he’s handwaving away?) Hope he has a lot of grad students…(glad I’m not one).
I, for one, look forward to my robot synthesis supervisor.
The infinite total synthesis theorem states that a postdoc hitting keys at random on a synthesis machine for an infinite amount of time will almost surely produce a given molecule, such as maitotoxin and a CETP inhibitor that doesn’t double as a lubricant due to excessive trifluoromethyl groups.
40,000 post-Docs doing reactions randomly will eventually produce Hamletomycin-D.
I’m not a chemist, but I am an embedded systems software guy (avionics and control systems). I think the software in something like this is going to have an 8- or 9-digit cost before we get it right.
How many “Things I Won’t Work With” could something like this generate per hour? Do we know enough to characterize these compounds, and can software be written to recognize them?
There are a large number of expensive mistakes to be made here.
How much money will be spent/spoilt on these projects ?
How many compounds would have been synthesized with this money ?
About the software itself. Since a couple of years now, people are talking about “big data”. In a word, algorithms cross data from scientific papers and find some unexpected relationship between things in fields far from each others.
Perhaps, this new technologie would help the design of this machine.
Tim Jamison and his group are doing a fantastic and beautiful chemistry ! Thank you for this link !
I’m with Derek and EDM.
I think that this will be quite the endeavor, in terms of a lot of things, time, money, effort, etc., but we’ll eventually get this kind of thing to a state where it’s significantly useful.
Here’s a better way to make billions of more novel compounds: Put several tons of household waste in a big blender, cook it for a few hours, and stick the goo down a massive HPLC.
The molecular printer is coming closer! Do you folks remember that vision, I believe Derek wrote about it in April-May this year.
I certainly would like to have even just a machine to more safely do reagent dispensing. Current methods take years of practice and are very much a craft. But, there have been other businesses that were artisan-run in the past but are industrialized now.
Just start by standardizing a flask and some sort of an air-free cannula for robotic handling, and that solves half of the problems. That has happened already in HPLC/GC sampling, so it is well within the current capabilities of robotics. Even more complex tasks are being automated today. For instance, a company called ZenRobotics sells robots that sort construction waste, a material stream that has no predictable size, shape or composition, and requires some AI heavy-lifting to understand and to derive the commands to the robot arm.
Unfortunately, the most obvious problem with this is that there no need, and no profit to be made. With overpopulation, there is no shortage of labor today, even skilled labor. And continuing “Asianization” means that this oversupply will increase. And you’ll get a 99% yield, both overall yield and in every step!
This one sounds like a tough ask… I think the quote “Nothing is impossible for the person who doesn’t have to do it” could be relevant here…
BTW, I still haven’t seen the robot that notices that reaction number 73 has a big clump of oil on the inside of the vial, and scratches it to see whether or not it crystallizes…
The stuck or clogged Pipeline problem has returned for me. Pages do not load reliably. Expanding from the Main Page (all topics) to see the Comments on a specific topic loads a blank page (with the proper URL in the address bar). Anyone else? HELP!
I think we will see robot chemists making compounds in the very near future. There are tons of reactions that ANYONE can do with a few days of training. A robot will easily handle this task, make the same compound with consistent yields (unlike humans) and less waste.
Nobody will have to worry about those chemists with “magic hands.” This is good for reproducibility too.
The devil, as they say, is in the details. It may well be possible, but from possible too useful is a big step.
In physiology, automated patch-clamp robots have actually existed for a while now, but they seem to have made very little headway in replacing humans.
“How about a cook machine that allows me to dial up a Kung Pao Chicken, Prof? Isn’t that more useful than the fancy schmancy Maitotoxin?”
Anonymous has it exactly right, even though the comment may have been intended as humorous. As long as there isn’t a robot to substitute a chef at a Chinese restaurant, which is a business in much more need of this technology, all this is bullshit. And so is the fate of any zloty and US dollars thrown into this black hole (unless they are used in other projects).
Imagine how the automated cook machine for a Chinese menu would work and you see how much more difficult it is for a synthetic project. And make no mistake about it, the Chinese restaurant robot cook will be here long before the robot chemist. It’s all well and good saying that this is technically feasible and it is possible in the future (a view that I agree with). It doesn’t really make me forget that in the here and now you’re dealing with some huge egos and an egregious example of misspent funding.
maybe they mean a 3-d printer that will punch out little plastic molecule versions of the real thing.
I’d be happy if they could just build a robot that found or ordered the chemicals that I need to run the reactions. Just to supply the solvents and chemicals for a big robochemist would take a vast number of people.
I just imagine what happens when the quench of a reaction in the robot goes bad and pops the septa, just like one of mine did recently. I was able to reseat the septa, cool it and continue with the reaction, but a mindless robot might not notice until it either explodes or the reaction turns to sludge. Either way would take a month to clean and fix, I would guess.
@ exGlaxoid: I am quite sure that Professor Whitby has thought about the issue and landed a solution. Imagine a sweet voice, like Siri or Cortana, calmly sound: “Alert one, alert one, Septa pops, reaction overflowing….you have ten seconds to leave the deck, silent count down…”
So how is this different than in silico toxicology? (see Derek’s post a couple of days ago). As we all know, we don’t need any biological input in tox, a computer can do it all. We don’t need a chemist’s input either, let the computer do it all for us.
On a side note, will it go bang if you tell it to string a bunch of nitrogens together in pretty ways? Sounds like a terrorist’s dream come true (if only they can modularize it small enough to fit in a backpack).
On second thought, it is also a drug dealers dream come true. Automated meth production, oxycodone, THC, cocaine, LSD, ecstasy, etc. You think of it, it’ll make it for you. Imagine a warehouse full of those chugging along.
^ If it works at doing what you want it to, it’s your dream come true, whatever it is that you’re doing.
Wouldn’t it be easier to clone Phil Baran
It does not matter if it works or not. What does matter is that breathless sales reps will convince nonchemist middle managers that it does – and real chemists will be let go, leaving kids fresh out of college to try and fix the damn robochemist every day and write reports about why it isn’t doing the job the sales guy promised.
fix the blog please..
Those who had a chance to see ArQule back in the days when combi-chem was the flavor of the month would realize that there is technology to potentially do this. I always thought that ArQule made a huge mistake by not keeping the service side of their business going while spinning off the “I wanna be a drug company” side but who am I to say.
Nonetheless, the cost will be huge as will the equipment. This is a long time off.
For this machine to be useful it does not have to make anything; merely ELIMINATE possibilities. If I need to make a particular Ar-Ar bond, and I can rule out Suzuki, Stille, Negishi, Buchwald, and Hartwig reactions, then that frees up resources to look at C-H activation or other routes for example. Such a technology will be VERY useful for that reason alone.
“Throw in the problem of figuring out a good synthetic plan (among the vast number of reaction and reagent possibilities) and the problem gets wildly, exponentially more hairy. ”
If this was the main problem, then it would not be a matter of time but funding to make this machine a reality. Machines are already better than humans at strategic thinking (http://en.wikipedia.org/wiki/Human%E2%80%93computer_chess_matches). However, this part of the project would need to be led by computer scientists, not a chemist.
These project can help make research much faster a all new robo chemist its gone be great project combining technology and chemistry together