AI-Generated Virtual Libraries Could Help Uncover RNA-Specific Regions of Chemical Space

Abstract

RNAs can recognize small-molecule ligands. However, the extent to which the molecules that they recognize differ from those recognized by proteins remains an open question. Cheminformatics analysis of experimentally validated RNA binders strongly suggests that RNA binders occupy a specific region of chemical space. However, less than 100 validated small molecule ligands are currently known. Here, we demonstrate how structure-based approaches could be used to navigate vast regions of the chemical space specific to ligand binding sites in five highly-structured RNAs. Our method involves using generative-AI to design target- and site-specific virtual libraries and then analyzing them using similar cheminformatics approaches as those used to assess experimentally validated RNA binders. Despite employing a completely orthogonal strategy, our results essentially reproduce the trends observed by analyzing the experimentally validated RNA binders. Large-scale generation of target and site-specific libraries may therefore prove to be helpful in simultaneously mapping the regions of chemical space unique to RNA and generating libraries that could be mined to identify novel RNA binders.