Myths of high-throughput experimentation and automation in chemistry

High-throughput experimentation (HTE) can be a powerful tool in chemical research, and the large datasets it generates could play role applying machine-learning methods to predictions. But high-throughput sometimes viewed as inaccessible academic chemist. This Voices, conjunction with our recent panel (https://www.brighttalk.com/webcast/16651/499811), asks: what are myths associated experimentation? What is of automation chemistry moving forward? I think biggest myth HTE that it’s mindless screening or just combinatorial chemistry. complete fiction. To use its full capacity, you need more about idea not less doing. There lot skeptics out there, naturally, when something new comes along. Some people dumbing down synthesis, but would say absolutely opposite. You really have you’re doing much than conventional batch-style Not all automatable. be, at end day, if get machine do routine work, means put your mind interesting things. If wanted weigh out, manually, 0.5 mg palladium acetate across 96-well plate, that’s mind-numbingly boring. for accurately, then I’m that. And my students their time productively. challenge getting breadth work on automated platforms. Certainly, experience, there engineering problems different chemistries scales, particularly those reactions partially heterogeneous. Even within area automation, there’s intellectual challenges solved we want fully synthetic organic need. still goal should looking for, because save so time, quality dataset generated by will reliable terms bigger picture data-driven science. The throwing bunch stuff against wall hoping positive result very myth. In fact, ask who first embarking HTE, “what's design philosophy”? Fundamentally, must hypothesis way exploring wide space, whether Bayesian optimization simple statistical techniques. While Bristol-Myers Squibb, group ran Mitsunobu reaction 180,000 possible configurations variables. By approach explored Doyle Adams groups Princeton, were able find perfect conditions only ~50 experiments. That’s 0.02% combinations. we’ll see coming years. These advanced learning techniques fundamentally changing efficiency which explore space using HTE. With data quality. Would rather 100 experiments run chemists, plate robot? variance lower latter results an internally consistent, high-fidelity dataset. For purposes, critical importance learn from transformation. Divorcing execution analysis major future direction, too. already companies cloud-based labs; they’ve essentially offloaded component, Chemists increasingly become scientists, pandemic certainly has highlighted well. As someone started off process chemistry, wasn’t believer first, everything micro-scale. start wonder, this milligram quantities, kilo? Is going scale? here tell does. I’ve seen again. HTE: observe one-milliliter vial one-milligram scale never lead real actually develop. It’s been busted many times. also acquire data. know, go home, potentially keep sample points, some kinetics, next beautiful analyzed, without standing over constantly taking samples. That good thing bad thing. More typically great; however, becomes bottleneck. believe accelerate labs, researchers set up robots while acquiring analyzing comfort homes. A younger scientists reactions, you’ll result. When was grad student, might count think, “Oh, five yesterday, did better labmates.” starting focusing weighed me didn’t saying, “I 96 great.” easily 1,500 fail—and time. Whether one 96, hard pick. beauty systematic capture. We’ve talked intelligence aspect this, having systematically captured no secret community done steal tools biochemistry automation. love point invented 1950, since then, biochemists enjoyed evolution infrastructure around well experimental tool. Synthetic chemists other early practitioners had compromise plates great solvents, air sensitivity stirring, Many true automate process, increase throughput, requires effort thought pull off.