(nanowork news) Japanese researchers have developed a machine learning process that simultaneously designs new molecules and suggests chemical reactions to make them. A research team at the Institute of Statistical Mathematics (ISM) in Tokyo published the results in a journal. Science and Technology of Advanced Materials: Techniques (“A Bayesian Approach for Simultaneously Designing Molecular and Synthetic Reaction Networks”).
Although many research groups have made great strides in designing feasible molecular structures with desirable properties using artificial intelligence (AI) and machine learning techniques, progress in the practice of design concepts remains limited. it’s slow. The biggest hurdle is the technical difficulty of finding chemical reactions that can produce designed molecules with viable efficiencies and costs for practical applications.
“Our new machine learning algorithms and associated software systems can design molecules with any desired property and suggest synthetic routes to create them from an extensive list of commercially available compounds,” said the research group leader. Statistical mathematician Ryo Yoshida said,
This process uses a statistical approach called Bayesian inference that works with huge data sets on different options for starting materials and reaction pathways. All possible starting materials are combinations of millions of compounds that are readily available for purchase. Computer algorithms evaluate a vast range of possible reactions and reaction networks to discover synthetic routes to compounds with directed properties to target. A professional chemist can review the results to test and refine what the AI suggests. AI makes suggestions and humans decide which one is the best.
“In a case study of designing drug-like molecules, this method showed overwhelming performance,” says Yoshida. We have also designed a route to industrially useful lubricant molecules.
“We hope our research will accelerate the process of data-driven discovery of a wide range of new materials,” concludes Yoshida. To support this purpose, the ISM team has made the software implementing the machine learning system available to all researchers on his GitHub website.
Current successes are focused solely on small molecule design. The research team now plans to investigate how to apply this procedure to polymer design. Many of the most important industrial and biological compounds are polymers, but creating new versions suggested by machine learning has proven difficult because finding reactions to build designs is difficult. I’m here. The simultaneous design and reaction discovery options offered by this new technology have the potential to break through that barrier.
