MolFoundry: A Structure-Aware Generative Framework for De Novo Design of PRRSV Nucleocapsid Candidate Binders

despite being one of the most economically devastating swine pathogens, porcine reproductive and respiratory syndrome virus (PRRSV) currently lacks approved antiviral therapeutics. The viral nucleocapsid (N) protein, being structurally conserved and indispensable for genome packaging, presents an attractive yet underexplored target for small-molecule inhibitor design. However, contemporary generative frameworks are predominantly structure-agnostic, frequently failing to reconcile geometric complementarity with drug-likeness and synthetic accessibility.

here, we present MolFoundry, a structure-aware de novo molecular design framework. MolFoundry integrates an SE(3)-equivariant pocket-ligand affinity scorer, a pocket-conditioned cross-attention generator, and a deterministic multiobjective optimization module guided by Pareto ranking and medicinal chemistry constraints. Trained on a comprehensive data set of 18,412 PDBbind complexes and specifically deployed against the PRRSV N-protein pocket, the framework generates highly diverse, synthetically tractable molecules with superior geometric consistency. Crucially, MolFoundry outperforms state-of-the-art diffusion, transformer, and reinforcement learning baselines across key metrics, achieving exceptional validity (0.938), perfect novelty (1.000), robust internal diversity (0.918), and highly favorable predicted binding affinity (-9.201 kcal/mol). Comprehensive ablation studies further substantiate that explicit pocket conditioning and multiobjective optimization are paramount for securing balanced pharmacological and affinity profiles. Data and software availability: MolFoundry is a Python-based molecular generation framework. The project code and documentation are available in the GitHub repository at https://github.com/hjd20030114-blip/MolFoundry.git under the MIT License.