RNAs adopt specific structures to perform their activities, beginning with transcription and ending with turnover. Conformational transitions among alternative structures are critical to virtually all RNA-mediated processes ranging from splicing to viral replication in eukaryotes. Understanding the numerous functions that RNAs play in living cells depends on informative three-dimensional structures. However, it is difficult to assess 3D structures of RNAs experimentally. Structures determined by NMR spectroscopy and X-ray crystallography are limited by the size or structural flexibility of the RNA.

We developed RNAComposer – a machine translation-based method for automated prediction of RNA three-dimensional structure from a user-defined secondary structure model. The method was evaluated using a representative benchmark set of 40 RNAs with different complexities. The quality of the input 2D structure is critical to obtain an accurate 3D structure. The in silico prediction of the RNA secondary structure is advanced and can be strengthened by incorporating constraints from chemical probing. However, defining precise secondary structures for 3D structure prediction remains challenging. Namely, multiple RNA conformations, pseudoknots and long-range interactions are important to evaluate. Furthermore, biologically relevant structures should be verified by independent analyses. During the presentation, I will introduce the RNAComposer method and discuss approaches we are developing to address difficulties arising from the prediction of 3D structures of large RNAs.