Sharon Aviran

RNA structure is a key determinant of RNA function and regulation. The coupling of chemical probing technologies, such as SHAPE, with deep sequencing has enabled large-scale experimental characterization of RNA structures in complex samples and under diverse conditions. Furthermore, probing data are often used to guide thermodynamics-based secondary structure prediction algorithms and have been shown to improve their accuracy. However, current algorithms treat these single-nucleotide measurements as statistically independent signals, inherently overlooking short-range dependencies in the data. Here, we show that discretized SHAPE data display context dependence within loop regions and within stem regions and we use Markov models to formally capture such dependencies. We then leverage Markov modeling in the classification of small structure motifs from their discretized SHAPE data signatures and subsequently integrate the classifying feature into the dynamic programming recursions that underlie computational RNA folding. Compared to state-of-the-art SHAPE-guided structure prediction methods, our Markov-informed framework improves prediction performance. Furthermore, we identify SHAPE signatures characteristic of highly stable hairpins, such as GAAA, GCAA, and UUCG tetraloops, and integrate these insights into the folding recursions to further improve predictions. Overall, the proposed framework provides a foundation for context-aware statistical modeling of SHAPE data, particularly in loop regions, where signal characterization has proven challenging due to high variance. This work further demonstrates that finer modeling of SHAPE data has the potential to push the limits of data-guided secondary structure prediction.

Nearest neighbor parameters are widely used in software for estimating the conformational stability of an RNA sequence folding into a specific structure. Folding stability for RNA with canonical nucleotides A, C, G, and U has been widely studied, but the same is not true for most modified nucleotides. In this work, we present a comprehensive set of nearest neighbor parameters for estimating the folding stability of RNA including pseudouridine in helical or loop contexts. These parameters are derived from 210 optical melting experiments involving helices with pseudouridine-A and pseudouridine-G pairs and with pseudouridine in loop motifs. The experiments include sequences with pseudouridine and U in the same strand, including U-A and U-G pairs, allowing us to consider the folding stability of sequences with both U and pseudouridine. On average, pseudouridine stabilizes RNA folding compared to U in an analogous motif, although this effect is sequence-context dependent. These parameters improve the modeling of folding stability for RNA secondary structures containing pseudouridine. We demonstrate that these parameters successfully model the secondary structure change for Saccharomyces cerevisiae U2 snRNA when two additional inducible pseudouridines are present. These parameters are freely available and incorporated into the RNAstructure software package.

The diverse regulatory functions, protein production capacity, and stability of natural and synthetic RNAs are closely tied to their ability to fold into intricate structures. Determining RNA structure is thus fundamental to RNA biology and bioengineering. Among existing approaches to structure determination, computational secondary structure prediction offers a rapid and low-cost strategy and is thus widely used, especially when seeking to identify functional RNA elements in large transcriptomes or screen massive libraries of novel designs. While traditional approaches rely on detailed measurements of folding energetics and/or probabilistic modeling of structural data, recent years have witnessed a surge in deep learning methods, inspired by their tremendous success in protein structure prediction. However, the limited diversity and volume of known RNA structures can impede their ability to accurately predict structures markedly different from the ones they have seen. This is known as the generalization gap and currently poses a major barrier to progress in the field. In this Perspective article, we gauge method generalizability using a new benchmark data set of structured RNAs we curated from the Protein Data Bank. We also discuss the emergence of deep learning methods for predicting structure probing data and use a new data set to underscore generalization challenges unique to this domain along with directions for future improvement. Expanding beyond improving predictive accuracy, we review how advances in deep learning have recently enabled scalable and accessible optimization of traditional structure prediction methods and their seamless integration with modern neural networks.

The diverse regulatory functions, protein production capacity, and stability of natural and synthetic RNAs are closely tied to their ability to fold into intricate structures. Determining RNA structure is thus fundamental to RNA biology and bioengineering. Among existing approaches to structure determination, computational secondary structure prediction offers a rapid and low-cost strategy and is thus widely used, especially when seeking to identify functional RNA elements in large transcriptomes or screen massive libraries of novel designs. While traditional approaches rely on detailed measurements of folding energetics and/or probabilistic modeling of structural data, recent years have witnessed a surge in deep learning methods, inspired by their tremendous success in protein structure prediction. However, the limited diversity and volume of known RNA structures can impede their ability to accurately predict structures markedly different from the ones they have seen. This is known as the generalization gap and currently poses a major barrier to progress in the field. In this Perspective article, we gauge method generalizability using a new benchmark dataset of structured RNAs we curated from the Protein Data Bank. We also discuss the emergence of deep learning methods for predicting structure probing data and use a new dataset to underscore generalization challenges unique to this domain along with directions for future improvement. Expanding beyond improving predictive accuracy, we review how advances in deep learning have recently enabled scalable and accessible optimization of traditional structure prediction methods and their seamless integration with modern neural networks.

RNA-binding proteins (RBP) play diverse roles in mRNA processing and function. However, from thousands of RBPs encoded in the human genome, a detailed molecular understanding of their interactions with RNA is available only for a small fraction. In most cases, our knowledge of the combination of RNA sequence and structure required for specific RBP-binding is insufficient for accurately predicting binding sites transcriptome-wide. In this context, the rapidly expanding collection of transcriptomic datasets that map distinct, yet intertwined post-transcriptional marks, such as RNA structure and RBP binding, presents an opportunity for integrative analysis to better characterize RBP binding. A grand challenge faced by our community is that relatively little information on the secondary structure context within and near RBP binding sites has been gleaned from integrating such datasets, partially due to lack of suitable computational methods. To engage scientists from diverse backgrounds in addressing this gap, the RNA Society organized the RBP Footprint Grand Challenge in 2021, an international community effort to develop new methods or leverage existing ones for predicting RBP binding sites through analysis of a growing volume of sequence, structure, and binding data and to experimentally validate select predictions. Here, we report the initiative, analyses and methods developed by the participants, validation results, and five new in vivo binding datasets generated for validation. We hope our work will inspire additional innovation in computational methods, further utilization of available data resources, and future endeavors to engage the community in collaborating towards closing other critical data-analysis gaps.

Protein synthesis in natural cells involves intricate interactions between chemical environments, protein-protein interactions, and protein machinery. Replicating such interactions in artificial and cell-free environments can control the precision of protein synthesis, elucidate complex cellular mechanisms, create synthetic cells, and discover new therapeutics. Yet, creating artificial synthesis environments, particularly for membrane proteins, is challenging due to the poorly defined chemical-protein-lipid interactions. Here, we introduce MEMPLEX (Membrane Protein Learning and Expression), which utilizes machine learning and a fluorescent reporter to rapidly design artificial synthesis environments of membrane proteins. MEMPLEX generates over 20,000 different artificial chemical-protein environments spanning 28 membrane proteins. It captures the interdependent impact of lipid types, chemical environments, chaperone proteins, and protein structures on membrane protein synthesis. As a result, MEMPLEX creates new artificial environments that successfully synthesize membrane proteins of broad interest but previously intractable. In addition, we identify a quantitative metric, based on the hydrophobicity of the membrane-contacting amino acids, that predicts membrane protein synthesis in artificial environments. Our work allows others to rapidly study and resolve the “dark” proteome using predictive generation of artificial chemical-protein environments. Furthermore, the results represent a new frontier in artificial intelligence-guided approaches to creating synthetic environments for protein synthesis.

RNA molecules are known to fold into specific structures which often play a central role in their functions and regulation. In silico folding of RNA transcripts, especially when assisted with structure profiling (SP) data, is capable of accurately elucidating relevant structural conformations. However, such methods scale poorly to the swaths of SP data generated by transcriptome-wide experiments, which are becoming more commonplace and advancing our understanding of RNA structure and its regulation at global and local levels. This has created a need for tools capable of rapidly deriving structural assessments from SP data in a scalable manner. One such tool we previously introduced that aims to process such data is patteRNA, a statistical learning algorithm capable of rapidly mining big SP datasets for structural elements. Here, we present a reformulation of patteRNA’s pattern recognition scheme that sees significantly improved precision without major compromises to computational overhead. Specifically, we developed a data-driven logistic classifier which interprets patteRNA’s statistical characterizations of SP data in addition to local sequence properties as measured with a nearest neighbour thermodynamic model. Application of the classifier to human structurome data reveals a marked association between detected stem-loops and RNA binding protein (RBP) footprints. The results of our application demonstrate that upwards of 30% of RBP footprints occur within loops of stable stem-loop elements. Overall, our work arrives at a rapid and accurate method for automatically detecting families of RNA structure motifs and demonstrates the functional relevance of identifying them transcriptome-wide.

The functions of RNA are often tied to its structure, hence analyzing structure is of significant interest when studying cellular processes. Recently, large-scale structure probing (SP) studies have enabled assessment of global structure-function relationships via standard data summarizations or local folding. Here, we approach structure quantification from a hairpin-centric perspective where putative hairpins are identified in SP datasets and used as a means to capture local structural effects. This has the advantage of rapid processing of big (e.g. transcriptome-wide) data as RNA folding is circumvented, yet it captures more information than simple data summarizations. We reformulate a statistical learning algorithm we previously developed to significantly improve precision of hairpin detection, then introduce a novel nucleotide-wise measure, termed the hairpin-derived structure level (HDSL), which captures local structuredness by accounting for the presence of likely hairpin elements. Applying HDSL to data from recent studies recapitulates, strengthens and expands on their findings which were obtained by more comprehensive folding algorithms, yet our analyses are orders of magnitude faster. These results demonstrate that hairpin detection is a promising avenue for global and rapid structure-function analysis, furthering our understanding of RNA biology and the principal features which drive biological insights from SP data.