New Method for Targeted Design of Molecules With AI
Complete the form below to unlock access to ALL audio articles.
Engineering novel molecules and materials with specific properties can yield significant advances for industrial processes, drug discovery and optoelectronics. However, the search for novel molecules and materials is comparable to looking for a needle in a haystack, since the number of molecules in chemical space is of the unimaginable order of 10 to the power of 60. That is significantly more molecules than there are stars in the known universe. Scientists at Leipzig University and the University of Warwick have now succeeded in developing a new method that enables the targeted design of molecules and materials with specific properties. They have now published their research findings in “Nature Computational Science”.
The researchers combined various artificial intelligence methods in their experiments. First author, assistant professor Julia Westermayr from Leipzig University’s Wilhelm Ostwald Institute of Physical and Theoretical Chemistry explains: “One model learned to predict quantum chemical properties of molecules, while the other learned how those molecules are constructed.” She adds that the first model is necessary to enable a high degree of accuracy when screening properties, as conventional methods for calculating quantum mechanical properties are highly time-consuming and require considerable computational power. In an iterative process that involves repeating steps until a certain target is reached or certain criteria are met, the researchers then used both models to generate new molecules and filter them according to certain properties. “In each round, the design model learned how the best-suited molecules are constructed and thus specifically predicted molecules with optimized properties in the next round,” says Westermayr.
Want more breaking news?
Subscribe to Technology Networks’ daily newsletter, delivering breaking science news straight to your inbox every day.Subscribe for FREE
The basis for the study was laid by Rhyan Barrett during an internship at the University of Warwick in England, funded by the Artificial Intelligence and Augmented Intelligence for Automated Investigations for Scientific Discovery (AI4SD) network.
Molecules with optimized properties
“We were particularly surprised that we were able to use artificial intelligence to find patterns in the data that led to optimized properties,” says Rhyan Barrett. Finally, the researchers managed to optimize multiple properties. This makes it possible to use the method to find Pareto-optimal solutions. A Pareto-optimal solution exists when the solution of several optimized properties has been found and the individual properties can only get better if another property gets worse in the process.
The method developed was used to design functional organic molecules relevant to optoelectronics. These novel, more efficient materials could be used in areas such as the solar energy industry, LED lighting, display technology, data storage, sensor technology and optical fibers in communications technology. The new method can also be transferred to other fields. Other potential areas of application include the development of active ingredients for new drugs with targeted, improved properties that are effective against specific diseases. Molecular design can also be used in environmental engineering to develop new processes for purifying waste water and air. In biotechnology, new biocatalysts and enzymes are developed based on the design of molecules with specific functions.
Reference: Westermayr J, Gilkes J, Barrett R, Maurer RJ. High-throughput property-driven generative design of functional organic molecules. Nat Comput Sci. 2023. doi: 10.1038/s43588-022-00391-1
This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source.