If you struggled in chemistry class, prepare to be blown away by the latest advancement in artificial intelligence.

A Harvard professor has created an A.I. system that can create new drug compounds. According to MIT Technology Review, scientists can program the system to come up with a new molecular structure that contains certain properties. For instance, a chemist can set parameters for something that is highly soluble and that can be used to treat pain. The system then uses its structural knowledge of more than 250,000 molecules and generates suggestions that combine those properties.

Pharmaceutical companies and researchers already use software for similar purposes, but the work has been limited since humans have had to come up with all the possible solutions. The new system uses deep learning, which means it becomes more intelligent as it gains more experience and studies more structures. It's similar to the way Google's image recognition software is better able to write captions as it views more photos, or how OpenAI learns to converse better by reading large scrolls of text.

Professor Alán Aspuru-Guzik, who invented the new A.I. program, tells MIT Technology Review that it won't replace chemists, but will act as a more advanced assistant than what they already have.

"It explores more intuitively, using chemical knowledge it learned, like a chemist would," Aspuru-Guzik says.

A.I. is playing a larger and larger role in science. Last month, the U.S. government announced a partnership with pharmaceutical company Berg Health to attempt to use A.I. to better diagnose and treat breast cancer. And startups like AtomWise and TwoXAR are using A.I. to predict what types of medicines will have desired effects and which will be toxic, and to suggest potential new uses for old drugs.

Aspuru-Guzik's software isn't quite ready for prime time--it still makes plenty of nonsensical suggestions. But his team is refining it, and hopes to soon feed it a trove of data on 100 million chemical structures that would give it an even deeper pool of knowledge.

Published on: Nov 8, 2016