AI in Action: Neural networks learn the art of chemical synthesis

See allHide authors and affiliations

Science  07 Jul 2017:
Vol. 357, Issue 6346, pp. 27
DOI: 10.1126/science.357.6346.27

You are currently viewing the summary.

View Full Text


Chemists looking to cook up new molecules face a challenge of choosing among hundreds of potential molecular building blocks and thousands of chemical reactions for linking them together. Computational chemists have long programmed computers with known chemical reactions, hoping to create software able to calculate successful molecular recipes. But reactions often don't work in a binary way, being either successful or not. Rather they produce a mix of products at different concentrations. So now researchers are looking to artificial intelligence for help. Instead of programming reactions as hard and fast rules, researchers have developed a neural network that learns from millions successful experiments and figures out on its own which reactions to choose to put together new molecules.