Computation and Structure‐Guided Arginine Scanning Engineers a Hyperactive AP Endonuclease for Multiplex Viral RNA Sensing

DNA-binding proteins such as AP endonucleases are powerful scaffolds for biotechnology, but there is no general way to rewire their DNA-contact surfaces. Here we report ARGENT (AI-guided Arginine scanning Engine for Nucleic-acid Tuning), an interpretable framework that uses protein-DNA structures and homologous sequences to propose beneficial Arg substitutions and multi-residue charge patches. Applied to the APE1-DNA complex, ARGENT combines structural and evolutionary features into a residue-wise hotspot score, compressing 276 candidate positions into 9 sites for testing. Six single mutants increase AP-site cleavage, and the top-ranked triple mutant, APE1-Evo, connects three DNA-contact patches, boosts the single-turnover rate constant ∼4-fold relative to wild-type APE1, and preserves mismatch discrimination at the AP-opposite position. We embed APE1-Evo into an upgraded NAPTUNE-V2.0 architecture that couples AP-site cleavage to a multilayer PfAgo cascade, enabling multiplex detection of dengue virus pseudotypes and direct typing of influenza A and B viruses in clinical RNA with high concordance to qPCR. ARGENT also transfers to other DNA-binding proteins, including MG34-1 Cas9d, highlighting its broader relevance to genome-editing enzymes. These results establish AI-guided Arg scanning as a practical route to engineer hyperactive, high-fidelity AP endonucleases and to tune DNA-binding interfaces for next-generation nucleic-acid technologies and viral RNA sensing.