Got a molecule in mind? Artificial intelligence-powered synthesis robots may soon take it from there. In a virtual event last week, IBM launched RoboRXN, a fully automated synthesis system. As project manager Teodoro Laino and his team discussed the platform, RoboRXN was in the background preparing a derivative of 3-bromobenzonitrile.
RoboRXN builds on IBM RXN for Chemistry, a free cloud-based software the firm launched in 2018 to predict the outcome of chemical reactions suggested by users. In 2019, Laino’s team added a retrosynthesis tool, enabling users to draw a molecule and have the software design a synthesis for it. Now, they’ve married that computational power with hardware, creating a robot that can execute the synthesis.
“With the RoboRXN technology, we are expecting to accelerate, profoundly, the way we do material discovery,” Laino said in a video accompanying the event.
Other groups have paired robots with AI-assisted synthesis planning, such as the the research nonprofit SRI International, which unveiled the SynFini system in January. What sets IBM’s system apart, Laino tells C&EN, is that the computer programs itself to carry the operation out, planning every injection, temperature change, shake, or swirl. It uses a routine he calls Smile2Actions, a reference to a chemical structure notation known as simplified molecular-input line-entry system, or SMILES. The system can also read text descriptions of synthetic methods and execute them (Nat. Commun. 2020, DOI: 10.1038/s41467-020-17266-6).
As of now, RoboRXN can handle up to five synthetic steps without human intervention. That covers a lot of molecular ground but could limit the system’s use on complex drug compounds. Extensive purifications, which are sometimes needed between synthetic steps, are also out of reach for the robot, though Laino says his team is working on it.
IBM hopes to offer RoboRXN on a fee-for-service basis soon and is looking for corporate partners to scale the system up. It plans to keep the reaction prediction and retrosynthesis tools available online for free.
Alán Aspuru-Guzik, a chemistry professor at the University of Toronto who studies machine learning, says IBM’s system is one of several that are bringing AI-powered synthesis closer to widespread adoption. “This is an important first step along the journey,” Aspuru-Guzik says. “Several groups in the world are working in the area and making rapid progress. Expect to hear from many, including us, in this space as well.”