AI-Powered Virtual Screening of Large Compound Libraries Leads to the Discovery of Novel Inhibitors of Sirtuin‑1

The discovery of new scaffolds and chemotypes via high-throughput
screening is tedious and resource intensive. Yet, there are millions
of small molecules commercially available, rendering comprehensive
in vitro tests intractable. We show how smart algorithms reduce large
screening collections to target-specific sets of just a few hundred
small molecules, allowing for a much faster and more cost-effective
hit discovery process. We showcase the application of this virtual
screening strategy by preselecting 434 compounds for Sirtuin-1 inhibition
from a library of 2.6 million compounds, corresponding to 0.02% of
the original library. Multistage in vitro validation ultimately confirmed
nine chemically novel inhibitors. When compared to a competitive benchmark
study for Sirtuin-1, our method shows a 12-fold higher hit rate. The
results demonstrate how AI-driven preselection from large screening
libraries allows for a massive reduction in the number of small molecules
to be tested in vitro while still retaining a large number of hits.