Integrated sensor engineering and AI-assisted peptide design to target the Zika virus NS2B-NS3 protease

Abstract Zika virus NS2B-NS3 protease (ZVP) is an essential enzyme for viral replication and a promising antiviral drug target. However, peptide-based inhibitor design remains challenging. Here, we developed an integrated workflow combining a recombinant FRET-based protease sensor with AI-driven peptide inhibitor design. The engineered PEprot ZIKV sensor, containing the ZVP cleavage sequence between two fluorescently labeled domains, enabled quantitative activity assays. PEprot ZIKV showed ~ 5.5-fold higher catalytic efficiency ( k cat / K m = 49 ± 7 × 10 6 M −1 s − 1 ) and a lower K m (4 nM) than a chemically synthesized substrate (22 nM) bearing the same cleavage sequence, with comparable catalytic turnover. Structural hotspot mapping of 53 ZVP crystal structures and multiple sequence alignment of flaviviral polyprotein motifs guided de novo peptide generation using RFdiffusion and ProteinMPNN. Computational filtering (docking, AlphaFold-3 co-folding, MD simulations and MM/PBSA) prioritized short peptides for testing. Biochemical screening with PEprot ZIKV identified three candidate inhibitors, with peptide 715 showing the strongest activity (IC 50 = 214 µM), followed by 630 and 701. Structure-guided peptide elongation improved 630-derived variants (e.g., 630_03, IC 50 = 224 µM) but reduced potency for 701- and 715-derived sequences. This integrated approach offers a generalizable framework to accelerate the discovery of peptide inhibitors against ZVP and other viral proteases.