Jerome Meneyrol
A system that uses artificial intelligence and a robotic arm to cook up new organic compounds could prove a big time-saver for chemists.
Historically, scientists trying to make a new drug or polymer have needed weeks to design, synthesize, test and refine a molecule. Klavs Jensen, Timothy Jamison and their colleagues at the Massachusetts Institute of Technology in Cambridge created a system that could develop a recipe for making desired molecules and then run the reactions to create them.
The team fed descriptions of millions of chemical reactions into a computer, which used the information to derive almost 164,000 rules describing the reactions that create particular molecular structures. The researchers then tasked an artificial neural network with choosing the rules that would produce a specific molecule of interest.
Next, a second neural network ran a simulation to determine whether that molecular synthesis was feasible. A human chemist refined the recipes before passing them to a robotic arm that moved various reactants into place and carried out the reactions. To demonstrate their invention’s talents, the researchers had it write and execute recipes for 15 common biomedical compounds.