A carregar...
Publicação
Evaluation of deep learning models for Aptamer Modelling in Silico
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.2 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Aptamers are single-stranded DNA or RNA oligo-nucleotides that fold into complex three-dimensional structures to bind molecular targets with high affinity and specificity (1). Often termed “chemical antibodies,” these biomaterials are vital for diverse applications in therapeutics, diagnostics, bioassays, and in vitro and in vivo imaging (2). Historically, aptamer discovery has relied on the Systematic Evolution of Ligands by EXponential Enrichment (SELEX), an in vitro process involving iterative selection, amplification and enrich-ment rounds. However, SELEX is time-consuming and often results in low success rates (3). Consequently, an in silico approach is essential to streamline the discov-ery process, reduce experimental costs, and provide insights into aptamer-target interactions. Recently, various deep learning models capable of predicting the 3d structure of aptamer-target complexes have become available. In this work, we evaluated gener-al biomolecular prediction models, like Alphafold-3, Protenix-v2, Chai-1, Boltz-2, Openfold-3, RoseTTA-Fold2NA, and two specialized aptamer modelling tools, AptaTrans and AptaBLE. Among the evaluated models, AlphaFold-3 stands as the current state-of-the-art general model for aptamer modelling, with Protenix-v2, Chai-1, Boltz-2 and RoseTTAFold2NA, accuracies falling short of AlphaFold-3. The only mod-el that rivals Alphafold-3 in aptamer-target bench-marks is OpenFold-3. On the other hand, specialized tools such as AptaTrans and AptaBLE, significantly outperform AlphaFold3 on validated DNA and RNA aptamers, which can be attributed to their optimiza-tion for sequence-based patterns. Overall, this work highlights the importance of an in silico approach for aptamer discovery and evaluates deep learning mod-els capable of predicting the 3D structure of aptam-er-target complexes. While general models provide a robust baseline for aptamer-target complex structure prediction, specialized tools offer enhanced perfor-mance for aptamer discovery. Future work will include a systematic comparison of the evaluated models.
Descrição
Palavras-chave
Aptamers Deep learning In silico
Contexto Educativo
Citação
Dias, F., Martins, F., & Sousa, S. F. (2026). Evaluation of deep learning models for Aptamer Modelling in Silico. Book of Abstracts of the 8th Meeting on Medicinal Biotechnology, 16. https://edicoes.ipp.pt/index.php/books/catalog/book/251
Editora
Politema