Here’s a useful guide to setting up lots of experiments – which is something that you’d think we science types already would know how to do, but not so. Michael Shevlin at Merck (NYSE:MRK) shares the experience of the labs there in reaction discovery and optimization over the years, and both urges people to try this mode of work and tells them to be cautious. He’s surely right on both counts – I think that properly run high-throughput work can be extraordinarily useful, but if you don’t pay attention to what you doing (and how you’re doing it), it can be a high-throughput waste of time. My own person Law of Automation applies: a robot is capable of screwing up an experiment faster and more thoroughly than you could ever screw it up by hand. It’s up to us humans to keep them from doing that.
In an era of declining resources and increasing demands in the research lab, there are several compelling reasons to con- sider HTE when conducting chemical research. First, novel synthetic approaches that require forming chemical bonds in unprecedented ways inevitably require significant amounts of experimentation in order to achieve breakthrough new discoveries. These tools allow a scientist to ‘go big’ and run orders of magnitude more chemistry than has been traditionally possible. Second, material limitations frequently restrict the breadth of conditions evaluated during a given step of a syn- thesis. The miniaturization inherent in HTE allows a scientist to ‘go small’ and run small arrays of experiments where limited amounts of precious scaffolds traditionally would have allowed only one or two experiments. Finally, many chemical transformations require routine reagent, solvent, and parameter screening in order to discover conditions that are good enough to push material forward to access the transformation of interest. HTE allows a scientist to ‘go fast’ and execute a single array of carefully chosen conditions in order to spend less time on routine synthesis and more time on research.
Just so. In my talk last fall on automation and AI in chemistry, I emphasized that last point, that the whole idea behind automation is to free us from grunt work. (The corollary is that, over time, more and more things gradually get defined as said grunt work). I think that eventually we may well end up with fewer people doing medicinal chemistry, but they’ll be working at, on average, a higher level than we tend to work at now, because they’ll be tasking with doing the things that the machines can’t do – and leaving everything else to the machines whenever possible. If you saw a chemistry department without autosamplers on their NMR or LC/MS machines, you’d be aghast at the waste of time and effort that this lack implies. More and more things will come to look just that odd, in time.
As another example of working at a higher level, Shevlin says: "Since HTE tools accelerate the execution of experimental arrays, we are afforded the luxury of spending more time carefully choosing their constituents." The advice is, basically, “measure twice, cut once” – spend your mental efforts in getting the most out of your experiments. For example, choose your solvent screen along several parameters (dielectric constant, functional groups, hydroxylic character, dipole moment), because these factors can work with and against each other in unexpected ways. Take advantage of your large number of experimental wells to be generous with control experiments in several dimensions. Spend some time thinking about how your various reagents are going to be dispensed, with attention to stability and reproducibility (the paper has advice on all these).
I remember trying to set up something along these lines when I was back in grad school, screening Lewis acids and other metal species on a carbohydrate-derived precursor, basically just to see what might form. The problem was, I had only TLC to tell me anything (no LC/MS!), and I had no 96-well plates and no multichannel pipetters to help me work through them even if I had. I liked the idea then, and I like it even more now that we have the tools to do such things more productively. High-throughput work can run from blue-sky “see what happens” experiments like that one, all the way to “what exact Pd catalyst conditions do I need for this specific transformation”. That covers a lot of ground, and because we all have a lot of ground to cover, I recommend giving it a try.