The authors demonstrated the application of generative artificial intelligence for creating new guest molecules for supramolecular systems. Described model utilizes three-dimensional representations of electron density and electrostatic potential to simulate interactions in host–guest systems. New guests are generated through optimization of shape and electrostatic interactions. These representations of generated guests are then converted into human-readable form using a transformer language model. The practical success of this approach was demonstrated with well-established molecular host systems, such as cucurbit[n]uril and metal–organic cages, resulting in the discovery of several previously unreported guests.
J. M. Parrilla-Gutiérrez, J. M. Granda, J. F. Ayme, M. D. Bajczyk, L. Wilbraham, L. Cronin, „Electron density-based GPT for optimization and suggestion of host–guest binders”, Nat. Comput. Sci.