How do you find new reactions? I blogged here in September about a very direct way of doing it, from John Hartwig's lab: set up a bunch of things and see what happens. I liked it very much, but opinions in the comments were mixed. Some people found this approach refreshing, while others found it more simplistic than simple.
Well, get ready for some more, courtesy of the MacMillan group at Princeton. This paper has just come out in Science on reaction discovery, and it takes a very similar approach to "accelerated serendipity". They were looking at photoredox catalysts, which have been used for some interesting studies in the past few years. You mostly see iridium and ruthenium catalysts, with variations of tris-bipyridyl ligands on them, but the variety of reactions that they can initiate is extraordinary.
Clearly, there must be a lot of reactions in this area that haven't even been found, and that's what this latest paper sets out to do:
Assuming that serendipity is governed by probability (and thereafter manageable by statistics), performing a large number of random chemical reactions must increase the chances of realizing a serendipitous outcome. However, the volume of reactions required to achieve serendipity in a repetitive fashion is likely unsuitable for traditional laboratory protocols that use singular experiments. Indeed, several combinatorial strategies have previously been used to identify singular chemical reactions (2–11); however, the use of substrate-tagging methods or large collections of substrate mixtures does not emulate the representative constituents of a traditional chemical reaction. On this basis, we posited that an automated, high-throughput method of reaction setup and execution, along with a rapid gas chromatography–mass spectrometry (GC-MS) assay using National Institute of Standards and Technology (NIST) mass spectral library software, might allow about 1000 random transformations to be performed and analyzed on a daily basis (by one experimentalist). Although we recognized that it is presently impossible to calculate the minimum number of experiments that must be performed to achieve “chance discoveries” on a regular basis, we presumed that 1000 daily experiments would be a substantial starting point.
That it would, and by combining a broad selection of interesting starting materials with several plausible photoredox catalysts, and then basically just letting things rip, they found one. Dicyanobenzene, as it turns out, does a radical coupling with tertiary amines, giving you a direct C-C bond formation route that arylates next to the nitrogen. It's a perfectly believable reaction, but there are a lot of perfectly believable reactions that you could draw in this area that don't actually work.
Looking over the paper, it appears that the more time-consuming parts of the experimental setup were avoiding known chemistry in the starting combinations, and looking over the results to see what was worth following up on in more detail. Those are both human-brainpower intensive tasks; the rest was automated as far as possible. Interestingly, it appears that MacMillan had earlier been trying a very similar approach to that Hartwig paper I blogged about in September, doing reaction discovery with transition metals. But they then switched to photochemistry, thinking that this might be a more wide-open field.
It's not like the reaction dropped out of the robotics fully formed. They saw a new product form with an iridium catalyst, dicyanobenzene, and N,N-dimethylaniline, but further optimization gave better (and more general) conditions. That's as it should be; there's no way (yet) to run enough experiments to both find new reactions and the best ways to run them in one shot. But just getting a whiff of something new and useful is enough, and I don't see any reason not to engage in automated searches for such things.
But from the reaction in the comments here to that Hartwig paper, I gather that not everyone agrees. As far as I can tell, one objection is that famous talented organic chemistry professors shouldn't have to engage in such brute-force exercises. The more elegant way to come up with these things, by this opinion, is to use more brainpower up front, rather than just mixing up a bunch of stuff to see what works. I suppose - not being a famous talented organic chemistry professor, myself - that I'm not so proud. But then, John Hartwig and Dave MacMillan are FTOCPs, and they seem to have swallowed their pride enough to find something new. And good for them!