Researchers from University of North Carolina at Chapel Hill (UNC) created an artificial-intelligence approach that is capable of designing new drug molecules from scratch.
The approach created by Alexander Tropsha, Olexandr Isayev and Mariya Popova, all of the UNC Eshelman School of Pharmacy, is called Reinforcement Learning for Structural Evolution (ReLeaSE). ReLeaSE is an algorithm and computer program that comprises two neural networks. One network, dubbed as ‘teacher’ provides the syntax and linguistic rules required for vocabulary of chemical structures. It contains a database of around 1.7 million known biologically active molecules. The other network, dubbed as ‘student’ is capable improving continuously on proposing molecules that develop effective medicines by learning through trial and error.
ReLeaSE effectively identifies viable drug candidates using virtual screening—a computational method widely used by the pharmaceutical industry. Although, virtual screening offers evaluation of existing large chemical libraries, the computational method only works for known chemicals. However, ReLeASE can create and evaluate new molecules and effectively generate generate molecules according to the required specifications. The molecules created can possess desired bioactivity and safety profiles. Moreover, molecules with customized physical properties such as melting point and solubility in water were created using the ReLeaSE method. The method was further capable of designing new compounds that facilitate inhibitory activity against an enzyme that is associated with leukemia. The research was published in the journal Science Advances on July 25, 2018.
“The ability of the algorithm to design new, and therefore immediately patentable, chemical entities with specific biological activities and optimal safety profiles should be highly attractive to an industry that is constantly searching for new approaches to shorten the time it takes to bring a new drug candidate to clinical trials,” said Alexander Tropsha.