Slashdot Mirror
AI Plus a Chemistry Robot Finds All the Reactions That Will Work (arstechnica.com)
A team of researchers at Glasgow University have built a robot that uses machine learning to run and analyze its own chemical reaction. The system is able to figure out every reaction that's possible from a given set of starting materials. Ars Technica reports: Most of its parts are dispersed through a fume hood, which ensures safe ventilation of any products that somehow escape the system. The upper right is a collection of tanks containing starting materials and pumps that send them into one of six reaction chambers, which can be operated in parallel. The outcomes of these reactions can then be sent on for analysis. Pumps can feed samples into an IR spectrometer, a mass spectrometer, and a compact NMR machine -- the latter being the only bit of equipment that didn't fit in the fume hood. Collectively, these can create a fingerprint of the molecules that occupy a reaction chamber. By comparing this to the fingerprint of the starting materials, it's possible to determine whether a chemical reaction took place and infer some things about its products.
All of that is a substitute for a chemist's hands, but it doesn't replace the brains that evaluate potential reactions. That's where a machine-learning algorithm comes in. The system was given a set of 72 reactions with known products and used those to generate predictions of the outcomes of further reactions. From there, it started choosing reactions at random from the remaining list of options and determining whether they, too, produced products. By the time the algorithm had sampled 10 percent of the total possible reactions, it was able to predict the outcome of untested reactions with more than 80-percent accuracy. And, since the earlier reactions it tested were chosen at random, the system wasn't biased by human expectations of what reactions would or wouldn't work. The research has been published in the journal Nature.
All of that is a substitute for a chemist's hands, but it doesn't replace the brains that evaluate potential reactions. That's where a machine-learning algorithm comes in. The system was given a set of 72 reactions with known products and used those to generate predictions of the outcomes of further reactions. From there, it started choosing reactions at random from the remaining list of options and determining whether they, too, produced products. By the time the algorithm had sampled 10 percent of the total possible reactions, it was able to predict the outcome of untested reactions with more than 80-percent accuracy. And, since the earlier reactions it tested were chosen at random, the system wasn't biased by human expectations of what reactions would or wouldn't work. The research has been published in the journal Nature.
And equally important: can it be networked to similar machines (preferably made by other manufacturers and run by different labs) to set up its own peer reviews?
And how soon before the drug cartels are buying up every machine that is produced to discover new substances?
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
A carefully selected group of these working in parallel could theoretically parse the entire possible set of reactions, given sufficient time. (Yes, I know that with infinite molecular weight, there is an infinite number of possible compounds. However, only so many heavy molecules are interesting or useful, and of those, there will be certain classes that are more interesting than others. This approach would permit investigation of pathways without actually expending reagents, once its models are accurate enough. That means after a certain amount of training, a theoretical molecule of interest could be presented to the AI, and it could shit out the ideal synthesis pathway, and the next efficient arbitrary "n" pathways.)
This is the kind of thing that is the beginning of universal replicators.
Be careful what you feed it.
"National Security is the chief cause of national insecurity." - Celine's First Law
It would have to learn how to do the chemistry at factory scales to do that. This is possible but unlikely to be economic.
If anything, the interesting work would be to go in the other direction. It is possible to make lab-on-a-chip units where the reagents are sealed in and each experiment is done in a small drop. This means you can repeat each experiment many times and see whether the results are repeatable. If your process is catalysed by some low concentration impurity, then that effect will vary with the number of molecules of that impurity, and you would expect a larger scatter in the yield.
First of all, most chemical synthesis reactions don't give one product with a 100% yield. Some molecules don't undergo the reaction, leaving some of the reactants behind. Some molecules have multiple sites where they can be chemically substituted, producing a statistical mixture of products. And some molecules have multiple isomeres that can be produced in the same reaction, but then behave differently in later reactions. So at each step of a chemical synthesis, you're polluting the system with more and more chemicals that might interact with each other and surprise you at later stages of the synthesis. The machine might perhaps be good at figuring out such interactions between biproducts of synthesis reactions even if a human just told it what the main product of each synthesis step would be.
This is a prime example of why patents are absurd and should be discontinued.
The skeptic in me wonders if 80% of the combinations had no reaction
Have gnu, will travel.