Small Single-Stranded DNA Structure Prediction

Abstract

Small single-stranded DNAs (ssDNAs) shorter than 100 bases, including antisense oligonucleotides and aptamers, have been extensively studied as promising therapeutics. Their three-dimensional (3D) structures can guide development and enhance clinical potential, but obtaining these structures is often costly and time-consuming. Only a limited number of small ssDNAs have their 3D structures experimentally determined. Therefore, there is a significant need to develop accurate computational tools to predict the 3D structures of small ssDNAs. In this work, we analyze two current strategies for ssDNA structure prediction: a traditional multistep approach and the AlphaFold 3 (AF3) strategy. We compare these strategies using a dataset of 36 ssDNAs, representing 7 unique structural motifs. Our results show that AF3 can be more efficient and accurate than the traditional strategy for predicting ssDNA structures, achieving a success rate of 56% (20 out of 36 cases). However, AF3 faces some challenges when predicting long ssDNA sequences or complex motifs, such as G-quadruplexes (G4) and junctions. Interestingly, molecular dynamics (MD) simulations can significantly improve the predicted structures generated by AF3. These findings offer valuable insights into the latest ssDNA structure prediction techniques and provide a practical guide for sampling ssDNA structures with AF3 and MD.