Automating synthesis from planning to execution | C&EN Global

Go to Volume 97, Issue 32. Volume 97, Issue 32. Aug 12, 2019 · Go to Volume 97, Issue 31. Volume 97, Issue 31. Aug 05, 2019 · Go to Volume 97, Issue 3...
1 downloads 0 Views 56KB Size
Concentrates Chemistry news from the week

▸ Highlights Isomers matter in some explosives CAR-Ts eliminate HIV in mice Catalysts convert racemates to single enantiomers Chemists find source of shark’s fluorescence DuPont says US government made it pollute Cambrex to be acquired and go private To lower mercury in fish, cut CO2 Nestlé India probed for research on preterm babies

6 8 9 10 13 14 16 17

SYNTHESIS

Downloaded via EAST CAROLINA UNIV on August 20, 2019 at 11:20:55 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Automating synthesis from planning to execution With a little human help, artificial intelligence and a robot find their way to drug molecules and advances in computer science have allowed humans to teach these rules to artificial intelligence algorithms, enabling them to plan chemical routes to target molecules. Concurrent progress in robotics and process chemistry similarly allow autonomous systems to follow directions and carry out multistep syntheses. The MIT researchers trained artificial intelligence algorithms on reactions drawn from US patents and the Reaxys database. Their synthesis planner uses those algorithms to propose synthetic routes, including reaction conditions, to a given molecule and to evaluate which path is best in terms of the number of steps and predicted yield. On the other side of the system, a robotic arm sets up experiments by connecting tubes supplying different reagents to flow-chemistry modArtificial intelligence plans chemical syntheses that ules like reactors and this robotic system can carry out—with some human membrane-based separahelp. tors. Other groups have demonstrated batch-chemistry-based The MIT team, which included chemautomation systems, but Jamison says ists, computer scientists, and chemical flow chemistry was better suited to auand mechanical engineers, drew on progress in two related areas: algorithmic pre- tomation for the types of reactions they were trying to do. It also enabled them diction of reactions and automated lab to run reactions at higher temperatures equipment. Chemical synthesis has proand pressures. The team’s robot could gressed from art to science over the centuries. Chemists like Robert B. Woodward synthesize compounds such as aspirin and Elias J. Corey helped develop the log- with a 91% yield and (S)-warfarin with a 78% yield and a 4.1:1 enantiomeric ratio. ical underpinnings of synthesis planning,

C R E D I T: CON N O R CO L EY A N D FE LI C E FRA N KE L

A new system uses artificial intelligence and robotics to carry out almost all the steps of chemical synthesis, from planning to execution, with just a little bit of human help (Science 2019, DOI: 10.1126/ science.aax1566). Timothy F. Jamison, Klavs F. Jensen, and colleagues at the Massachusetts Institute of Technology demonstrated the system could synthesize several drug molecules and two small compound libraries.

It also synthesized two libraries of five drug-related compounds each. Although other researchers have developed AI synthesis planners or automated synthesizers, the MIT group’s system comes closest to a fully automated system. “I have little doubt that this concept will dramatically accelerate synthetic design and will change the way we do synthesis. It is truly game changing,” says Cathleen Crudden, an organic chemist at Queen’s University in Ontario. Human chemists do play a role in the group’s system. They edit the recipe generated by the AI before it’s fed to the robotic chemist. “We still need a chemist to fill in the details” of precise concentrations, temperatures, and timing, says coauthor Connor Coley, lead developer of the system’s software. Jensen says the papers and patents that their AI learned from didn’t always include this information at the level of detail their algorithms need. But Coley says there’s no reason an algorithm couldn’t do this step as well if it had the right data. The group doesn’t want to replace human scientists. “The aim here is not to eliminate the chemist,” Jensen says, but to give chemists more time for creativity by automating routine synthetic tasks. Paul Richardson of Pfizer, who has explored automation, notes that past attempts at automating synthesis haven’t yielded much return on investment but says this system is an exciting advance. And he appreciates the researchers’ reliance on humans, adding, “It is refreshing to see the authors acknowledge the limitations of utilizing computational predictions as well as the need in some cases for expert human intervention.”—SAM

LEMONICK AUGUST 12, 2019 | CEN.ACS.ORG | C&EN

5