Prediction of RNA-RNA interactions

The majority of an organisms transcriptome does not consist of protein encoding RNA but represents so called non-coding RNA (ncRNA), most of them acting as regulatory elements. Many of these ncRNA molecules require interactions with protein-coding messenger RNAs or other ncRNAs to fulfill their regulatory functions. A detailed understanding of these RNA-RNA interactions gives insight into the regulatory networks of organisms. Due to the complexity and expense for experimental studies, there is a strong need for com- putational prediction approaches. Current methods are usually restricted to simple interaction types and show a low prediction accuracy. Within this project we will identify steric and topological features constraining known inter- actions to increase prediction quality of RNA-RNA interaction prediction models. Furthermore, we will define new models covering more general interaction patterns up to multi-site interac- tions. Identified features will in particular be integrated as hard constraints to avoid unrealistic structures. This will be accompanied with an extensive research of the kinetics of RNA-RNA interac- tions, since it is able to guide the interaction formation to non-predictable, suboptimal struc- tures. Therefore, new energy landscape models will be defined and implemented that allow for a detailed but still computationally accessible study of the energy landscape based kinet- ics. This includes the development of new methods for an efficient exploration of large energy landscapes as well as for the computation of transition model and kinetics. These two research fields are joined within new RNA-RNA interaction prediction pipelines that will respect the new structure constraints as well as account for kinetic effects defining the interaction formation. We will apply our new pipelines to investigate interaction networks for ncRNAs, to screen for targets of regulating ncRNAs, and to increase knowledge of the mechanistic details of certain binding pathways. One direction of application will be the investigation of the mechanistic details of anti-sense RNA interactions. Anti-sense RNAs are found in great extent in the transcriptome of pro- and eukaryotes while there is only limited knowledge about their regulatory impact. An especially interesting topic is to determine to what extent anti-sense transcripts actually form duplexes with their respective sense transcripts and what regulatory mechanisms are possible. Our computational studies will be supported by wet-lab experiments. Beside new insights, the project aims at the development of a large tool set to promote further research in this field. The targeted results do not only support RNA-RNA interaction studies but also enable progress for structure prediction of single molecules as well as kinetics studies of large energy landscapes of other systems.

Ribonucleic acids (RNAs) are best known as messenger RNAs (mRNAs) that act as blueprints for the production of proteins in the ribosome. However, the majority of RNAs in our cells do not code for proteins. These noncoding RNAs (ncRNAs) perform a variety of essential regulatory functions. More often than not, these functions depend on the fact that RNAs can recognize and bind other RNAs, forming RNA-RNA interactions e.g. between a regulatory ncRNA and its mRNA target, thus controlling the expression of a gene. The regulatory potential of RNAs has led to increasing interest in RNA therapeutics, as exemplified by the novel mRNA based vaccines used to combat the Covid-19 pandemic. Accurate prediction of RNA-RNA interactions is thus essential both for our understanding of cellular biology, as well as RNA based applications in biotechnology and medicine. In this project we were able to identify two major limitations in the existing approaches for RNA-RNA interaction prediction: (i) some predicted interaction structures are sterically impossible (ii) they are based on energy minimization and ignore folding kinetics Available tools for RNA-RNA interaction prediction all work on the level of secondary structure, i.e. they consider base pairs and helices formed between two RNAs without tackling the (hugely complex) problem of predicting the detailed 3D structure. While this is much more efficient, it means that some predicted secondary structures are impossible in the sense that they cannot be implemented as a tertiary structure. By modeling the formation of an important type of RNA-RNA interaction, the kissing hairpin, on the tertiary structure level, we were able to determine the maximum size of the interaction in kissing hairpins. This information can now be used to filter classical RNA-RNA interaction predictions and discard infeasible structures. In order to study kinetics we designed a simple model of RNA-RNA interaction formation that considers all folding pathways that lead from an initial contact between two RNA molecules to a fully formed interaction. The model can be used to compute a variety of features characterizing the interaction. For example, the probability that the two RNAs will fall apart before reaching the full interaction. Using a machine learning approach, these features were combined with classical RNA-RNA predictions to improve the accuracy of target predictions for bacterial small RNAs. The project also resulted in improvements in software tools that are widely used throughout the RNA research community, such as accuracy improvements in the IntaRNA tool developed by our collaboration partners in Freiburg, as well as the ability to predict interactions of multiple RNAs in the Vienna RNA package.