Deciphering Complex RNA structure by probing and predictions

RNA molecules are essential building blocks of life that fulfill important regulatory functions in our cells. The structures of these molecules play a particularly important role, as they determine how RNAs interact with each other and with other components of the cell, such as proteins. So if one knows the structure, one can infer the function. In this project, researchers from Austria and France are working on methods to decode these structures of RNAs. Experts from the fields of biology, biochemistry, structural biology, computer science and physics are developing new integrative approaches to discover the secrets of RNAs using experimental methods in the laboratory as well as computer programs.

To unravel and model the structure of a biomolecule is one of the most important aspects in understanding its nature and ultimately its function. This is not only important to identify their interaction with other molecules in the cell, but also key for novel targets and strategies to treat common human diseases such as cancer or viral infections. However, exact structures can typically only be obtained through time- and cost-intensive experiments, which in some cases cannot even be applied to specific biomolecules. In our project, we focused on the biopolymer RNA and aimed to shed light on an alternative to obtain its structure: computational RNA secondary structure predictions guided by experimental high-throughput RNA probing data. We carefully investigated the informative value of such data and, as a result, developed a new approach that combines multiple probing reagents and diverse environmental conditions. Moreover, we introduced a method to assist structure prediction using another source of high-throughput data, which has not previously been employed for this purpose. The most common approach to predict RNA secondary structures is physics-based. It takes nucleotide sequences as input and uses thermodynamic energy parameters to compute the most likely structure(s). This type of algorithm performs well for short RNA molecules, but often fails for longer RNAs due to the vast number of similarly stable structural alternatives and simplifying model assumptions. A substantial improvement in prediction accuracy can be achieved by incorporating energy parameters derived from RNA probing experiments. Probing data are typically reported as per-nucleotide reactivities, indicating either the RNA backbone flexibility or the accessibility of the nucleotides to the probing reagent, depending on the method used. In our studies, we found that relying on one of the most popular RNA probing methods of the past decade alone may easily mislead the structure modeling process. To mitigate the uncertainties inherent in such experiments, we proposed and developed a novel approach that exploits the complementarity of multiple datasets using different reagents and environmental conditions. In particular, variation in factors such as metal-ion concentration and temperature can help disentangle structural features that are detected experimentally but not represented in the prediction model, thus significantly enhancing prediction performance. Interactions between RNAs are typically explored through cross-linking experiments that produce sequencing reads composed of parts of distinct molecules. To aid secondary structure prediction, we instead developed a method that uses reads mapping to distant regions of the same RNA. This approach outperforms traditional probing techniques, as it provides two-dimensional information on intramolecular interactions critical for accurate structure modelling. By extending the repertoire of structure-informative data sources, our method paves the way for more accurate and scalable RNA structure analyses - an essential step toward understanding RNA roles in health and disease.