Physicochemical rule-embedded generative modeling for constrained peptide therapeutics

Constrained peptides couple covalent topology, stereochemistry, three-dimensional structure, conformational dynamics and developability. We review physicochemical rule embedding across chemical-language, target-conditioned structural, geometric, atom-bond, ensemble and property-guided generative models, distinguishing generated molecular objects from downstream inference and predictions from prospective evidence. We identify limitations in chemical coverage, applicability, uncertainty and pipeline attribution, and routes towards identity-preserving, environment-aware and experimentally calibrated design.