Small Single-Stranded DNA Structure Prediction

Small single-stranded DNAs (ssDNAs) under 100 bases, including antisense oligonucleotides and aptamers, have been widely explored as promising therapeutics. Their three-dimensional (3D) structures are essential for guiding the development and enhancing clinical applications. However, experimentally determining these structures is often costly and time-consuming, resulting in only a few small ssDNAs with known 3D structures. Consequently, there is a pressing need for accurate computational methods to predict ssDNA structures. In this work, we examine two current strategies for ssDNA structure prediction: a traditional multistep approach and the AlphaFold 3 (AF3) approach. We evaluate these strategies using a dataset of 149 ssDNAs without ligands, representing six distinct structural motifs. Our findings indicate that AF3 is often more efficient and accurate than the traditional method, achieving a success rate (GDT = 1) of 38% (57 out of 149). Nevertheless, AF3 encounters difficulties with longer sequences or complex motifs such as G-quadruplexes and junctions. Notably, molecular dynamics (MD) simulations can substantially enhance the AF3-generated structures. These insights shed light on the latest ssDNA prediction techniques and offer a practical guide for sampling ssDNA structures using AF3 and MD.