Predictive modeling and fragment-based design of triterpenoid-nucleoside conjugates targeting coronavirus membrane fusion.

Triterpenoids are known for their broad biological activities, yet their potential against SARS-CoV-2 has not been fully elucidated. In this study, using an antiviral dataset generated from our experimental evaluation of triterpenoid-nucleoside derivatives that disrupt SARS-CoV-2 pseudovirus entry by targeting the fusion peptide-heptad repeat 1 (FP-HR1) interface of Spike, we developed a stacking ensemble model with strong predictive performance (accuracy = 74.17 %, AUC = 0.85) to facilitate efficient active-compound identification. Guided by the machine-learning model and a fragment-based assembly strategy, we designed and prioritized a series of new derivatives, from which nine candidates exhibiting >80 % inhibition at 15 μM were identified. Subsequent analysis indicated that conjugates incorporating 2',3'-O-isopropylidene-protected adenosine or inosine moieties demonstrated the highest levels of anti-SARS-CoV-2 pseudovirus activity. These compounds exhibited micromolar EC50 values against SARS-CoV-2 wild-type (0.76-4.84 μM) and Omicron (1.94-5.18 μM) pseudovirus in BHK-21-hACE2 cells. Biophysical and biochemical analyses indicate that these conjugates interfere with viral entry by engaging a conserved fusion-region encompassing the FP-HR1 interface of the Spike S2 subunit. This work suggests that triterpenoid-nucleoside conjugates represent a promising starting point for the development of SARS-CoV-2 entry inhibitors and illustrates the utility of computational approaches for guiding antiviral compound discovery.