Discovery of Tetrahydroisoquinoline-Based SARS-CoV-2 Helicase Inhibitors with Iterative, Deep Learning-Enhanced Virtual Screening.

In this study, we pursued a structure-based drug discovery campaign targeting the SARS-CoV-2 helicase through three rounds of virtual screening (VS) enhanced with Artificial Intelligence (AI). The third round incorporated a deep neural network (DNN) in a virtual screening protocol to prioritize commercial molecules containing fragments related to previously identified hits. This model predicted binary activity toward the target and contributed to a 21% improvement in hit-identification efficiency compared with the previous approach, which used the same virtual screening protocol. In total, we have identified at least six hits with selectivity indexes (CC50/EC50) above three, which show promise in early stages of SARS-CoV-2 antiviral development. Additionally, a subfamily of 18 3-phenyl-1,2,3,4-tetrahydroisoquinoline (THIQ) derivatives was evaluated. The most potent compound identified in this series is MWAC-3429, which displayed cell-based antiviral activity (EC50 = 5.4 μM) with no observable cytotoxicity. The predicted binding site for this THIQ chemotype is located at the interface between the RecA1 and Stalk domains of Nsp13, where it appears to disrupt the 1B-RecA2 interdomain interactions, suggesting a novel allosteric inhibition mechanism, as supported by molecular dynamics simulations. Furthermore, five-membered ring substituents at the THIQ amino position were observed to enhance potency relative to their six-membered ring counterparts.