AI-driven biomolecular design: Modalities, models, and translation

The integration of artificial intelligence (AI) into biomolecular and materials science is reshaping the design, optimization, and translation of functional biomaterials. Although empirical screening and iterative experimentation remain indispensable, they are increasingly insufficient for exploring the high-dimensional sequence and structural spaces of biomolecular modalities. Peptides, antibodies, and aptamers are particularly important sequence-defined modalities because they can either form material architectures or endow biomaterial platforms with molecular recognition, responsiveness, and biological function. Their performance, however, depends not only on intrinsic molecular activity but also on whether folding, assembly, stability, and function are preserved after synthesis, conjugation, formulation, and integration into material systems. In this context, this review synthesizes advances from 2020 to early 2026 in AI-driven biomolecular design, with emphasis on peptides, antibodies, and aptamers as programmable components of therapeutic and diagnostic biomaterials. We first examine the advantages and modality-specific constraints of these systems. Thereafter, we analyze major predictive, generative, and optimization-based AI models relevant to their design. Finally, we make the case that closed-loop AI workflows, while not yet mature in biomaterial design, provide a practical framework for addressing downstream translational bottlenecks. By linking computational design with iterative synthesis, material integration, and biological validation, closed-loop workflows can improve translation by bringing downstream constraints into the design process itself. Through this model-modality-translation organization, the review provides a critical map for the AI-driven rational development of next-generation biomolecular biomaterials.