Summary
A bioRxiv preprint describes REAL-SWIT, a generative system for exploring make-on-demand chemical space. In experiments on ROCK1, six of 23 synthesized compounds inhibited the target in biochemical tests.
Researchers at Westlake University describe REAL-SWIT, a generative system for searching Enamine REAL Space, a large make-on-demand chemical space. In a bioRxiv preprint posted September 23, 2026, they report that the system identified six biochemical inhibitors of the target protein ROCK1 among 23 compounds they synthesized. One, RX-3, had a reported IC50 of 0.17 μM.
Virtual screening uses computational methods to rank molecules for possible activity against a biological target before laboratory testing. The researchers’ approach combines a model that generates complete molecules resembling those in REAL Space with a target-specific scoring model to guide docking. Docking estimates how a molecule might fit and interact with a target; its scores help prioritize candidates but are not themselves experimental measurements of activity.
Searching chemical space molecule by molecule
Chemical spaces can contain far more possible molecules than can feasibly be enumerated and docked one at a time. REAL-SWIT is designed to explore this space by generating candidate molecules directly, rather than building a search only from smaller fragments or components.
The team reports that approximately 96% of the generated molecules were present in REAL Space. Some generated molecules missing from the release used for training appeared in later, expanded releases. The authors interpret this as evidence that the model learned features of the space that extend beyond its training release.
In computational benchmarks, the researchers report that REAL-SWIT found more molecules with favorable docking scores than representative fragment-based search strategies. They also describe cooperative combinations of building blocks that fragment-level prioritization tended to miss. The result illustrates a potential advantage of scoring a complete candidate: combinations that are not promising when considered at the fragment level may receive favorable scores as part of a larger molecule.
ROCK1 compounds were tested in biochemical assays
To test whether the computational search could yield active molecules, the team synthesized 23 candidates for ROCK1 and found six biochemical inhibitors. RX-3 had an IC50 of 0.17 μM. IC50 is the concentration of a compound associated with 50% inhibition in a specified assay; the value depends on the assay conditions, so it is not by itself a measure of how a compound would perform as a medicine.
The biochemical results provide an experimental check on the screening approach, beyond its computational docking scores. They are early-stage evidence from tests against a molecular target: the preprint does not report these compounds as treatments or establish effects in animals or people. The study is posted as a preprint on bioRxiv.