Generative Access to Make-on-Demand Chemical Space Enables Ultra-Large Virtual Screening

Ultra-large-scale virtual screening (ULVS) has grown reliant on make-on-demand chemical spaces; however, efficiently accessing these spaces can be difficult because their size surpasses what can be exhaustively enumerated and docked. In this study, researchers introduce REAL-SWIT, a generative ULVS framework. This framework learns Enamine REAL Space as a synthesizable molecular distribution and integrates it with a target-specific scoring model.

This combination enables docking-guided exploration of REAL Space at the level of complete molecules. The generative model that underpins REAL-SWIT is capable of capturing transferable features from the REAL Space distribution. Remarkably, around 96% of molecules generated through this process were found within REAL Space. Moreover, a subset of molecules generated that were not initially included in the training release later appeared in subsequent expanded releases.

In computational benchmarks, REAL-SWIT outperformed fragment-based search strategies by identifying more molecules with favorable docking scores. This includes combinations of cooperative building blocks that fragment-level prioritization might have overlooked. To validate the approach experimentally, researchers screened ROCK1 and identified six biochemical inhibitors among the 23 synthesized compounds.

One such inhibitor, RX-3, exhibited an IC50 of 0.17 M. These findings demonstrate that generative access to make-on-demand chemical space serves as a practical strategy for ultra-large virtual screening.