In silico aptamer design: from sequence selection to structural optimization and computational modelling strategies.

Aptamers are of growing importance due to their high specificity, tunable affinity, and versatility across therapeutic, diagnostic, and biosensing applications. Unlike antibodies, they can be synthesized rapidly, modified easily, and remain stable under diverse conditions, making them powerful tools in both fundamental research and translational healthcare. The De Novo approach to aptamer development represents a paradigm shift from traditional SELEX-based methods toward fully computational, data-driven design. This strategy integrates bioinformatics, artificial intelligence, and molecular modeling to generate high-affinity aptamer sequences without relying on pre-existing libraries. Beginning with target characterization, ranging from small molecules to macromolecules and whole cells, the pipeline employs tools like SwissModel and AlphaFold for structural prediction, and FTsite or Fpocket for binding site identification. AI techniques such as Monte Carlo Tree Search, Hidden Markov Models, and Variational Autoencoders guide sequence generation, followed by secondary structure prediction and molecular dynamics simulations for interaction validation. Advanced platforms including AptaTRACE, RaptGen, AptaDiff, AptaTrans, and AiDTA demonstrate the power of generative modeling, transformer architectures, and reinforcement learning in aptamer discovery. These frameworks enable first engineering, validated through SPR and in vitro assays, and offer scalable solutions to overcome limitations in dataset size, structural modeling, and experimental throughput. The De Novo approach thus establishes a robust design-test-learn cycle, positioning AI-driven aptamer development as a cornerstone of next-generation diagnostics, therapeutics, and biosensing.