AI-designed PNA-peptide chimera overcomes suboptimal binding for dual inhibition of viral RdRp.

The chimera combining the peptide nucleic acids (PNAs) and peptides represent a promising bifunctional strategy by concurrently binding with protein catalytic pocket and its associated RNA template, effectively disrupting protein's function. Conventional designs face challenges due to non-optimized peptide-protein interactions and empirical PNA sequence selection, which lack thermodynamic or computational refinement, compromising selectivity and efficacy. Here, we present an artificial intelligence (AI) and molecular simulations-driven framework for the de novo design of a high-affinity PNA-peptide chimera targeting SARS-CoV-2 RNA-dependent RNA polymerase (RdRp) through synergistic inhibition of its catalytic pocket and RNA template. Leveraging a hybrid model trained on 2950 protein-protein and peptide-protein complexes, we first decoded residue-residue interaction propensities to rationally engineer a 5-mer peptide (LEU-VAL-SER-GLU-ASP) with optimized RdRp binding (ΔG = -9.89 kcal/mol). Concurrently, thermodynamic profiling yielded a 6-mer PNA (GAUUAA, ΔG = -11.85 kcal/mol) with high RNA complementarity. The structurally integrated chimera exhibited markedly enhanced binding affinity with Kd = 2.6 nM, 56 % lower than the peptide counterpart (Kd = 45 nM), and potent in vitro antiviral activity (IC50 = 9.10 μM, SI = 11.2) than the peptide (IC50 = 26.38 μM, SI = 4.4). The chimera displayed high specificity for SARS-CoV-2 RdRp with negligible cross-reactivity to SARS-CoV-1 and RSV homologs. This study demonstrates a potential framework for rational design of chimera with high specificity and inhibitory potential.