Computational & AI arXiv
RL-generated neuraminidase inhibitors show activity against influenza A and B in cells and mice
- A reinforcement-learning molecule design platform generated a large, diverse library aimed at influenza A and B neuraminidase, and most designs had better physicochemical profiles than oseltamivir.
- After computational filtering, a small set of non-sialic-acid-like compounds was tested experimentally; two inhibited both virus types, including an oseltamivir-resistant mutant, and simulations suggested oseltamivir-like residue contacts.
- Both hits were active in cell assays and gave protection in infected mice, supporting generative RL as a route to antivirals validated in vitro and in vivo.
Beating oseltamivir on physchem is a low bar, since it's a polar carboxylate prodrug with modest properties. The real test is whether the hits are truly novel versus known NA chemotypes (zanamivir, peramivir, other transition-state mimics), and whether RL beat a simple enumeration or docking baseline. Next step: enzyme IC50 shifts across H275Y, plus PK and dose-response in vivo.
Paper: arXiv ↗ Discovery of new antiviral agents through artificial intelligence: in vitro and in vivo results