Predicting single-stranded DNA oligonucleotides 3D structures: an open issue

Abstract

Single-stranded Nucleic Acids (ssNAs) play major biological functions and represent an interesting biotechnological tool. Their function depends strictly on the specific foldings they can adopt. Therefore, information about ssNAs tridimensional structures is fundamental to investigate their functions. In this context, in silico 3D structure prediction can facilitate ssNAs design. Many algorithms have been developed with this aim, mainly focused on RNA. However, the growing interest in single-stranded DNA (ssDNA), due to their greater stability as compared to RNA, has highlighted the need to adapt these methods for ssDNA. This study assessed three RNA 3D structure prediction methods (RNAComposer, SimRNA, and Vfold3D), based on their performance in the Critical Assessment of protein Structure Prediction 15 and/or 16, to evaluate their applicability to ssDNA. At this scope, a dataset of 93 experimentally determined ssDNA structures, including challenging motifs such as G-quadruplexes, was built. Various metrics, such as RMSD, GDT TS, and INF were employed to benchmark the accuracy of the predictions. The three tools showed similar and moderate performances. In addition, they show strong difficulties in modeling G-quadruplexes, and structures containing motifs strongly increasing the intrinsic flexibility of ssDNA. Despite the recent efforts in the prediction of the 3D folding of ssNAs, it is clear that a significant improvement of the methods is needed. This should involve taking into account the conformational variability of this kind of molecules and paying attention to their specific 3D motifs. Author summarySingle-stranded oligonucleotides (ssNAs) are RNA or single-stranded DNA molecules involved in crucial biological processes, such as gene expression, DNA replication, and transcription. In addition, they represent a powerful biotechnlogical tool exploitable in therapeutics, diagnostics and biosensing. This is due to their capacity of recognizing different kind of molecular targets, thanks to the tridimensional foldings they can adopt. It is therefore clear that the knowledge of the ssNAs structure is fundamental to master these molecules. With this aim, much effort has been paid in developing computational tools for the prediction of ssNAs 3D structures. However, their application is mostly limited to RNA sequences, even if the interest in ssDNAs is rapidly increasing. Moreover, an extensive benchmark on their performances is missing. We focused this work on assessing the performances of three ssNA structure prediction tools, which best performed in the two latest Critical Assessment of protein Structure Prediction contests, on a large dataset of ssDNA, in order to establish the strengths and limits of the available tools. This knowledge will be helpful in finding new solutions for better understanding the ssNAs folding and their structure-function relationship.