RNA 3D structure modeling dates to the late 1960s and several computer programs for predicting RNA 3D structures have been proposed since then. RNA-Puzzles is a collaborative effort dedicated to advancing and improving RNA 3D structure prediction. With the agreement of crystallographers, RNA structures are predicted by different groups before the publication of crystal structures. Since the success of AlphaFold in protein structure prediction, artificial intelligence approaches are continuously designed to solve the problem of RNA 3D structure prediction with strategies like AlphaFold. However, eliminating redundancy between training and test data is not trivial and some programs have shown overfitting results. Therefore, blind, unbiased evaluations (based on equivalence of comparison metrics) of all prediction tools are a necessary requirement.
A dedicated website ( http://www.rnapuzzles.org/) gathers the systematic protocols and parameters used for comparing models and crystal structures, all the data, analysis of the assessments, and related publications. Up to now, 40 RNA sequences with experimentally determined structures (x-ray or cryo-EM) have been predicted by many groups from several countries. Many of the predictions have achieved high accuracy after comparison with the solved structures.

Rfam is a database of over 4,000 non-coding RNA (ncRNA) families. Each family is composed of a sequence alignment called the seed, often manually curated, a consensus secondary structure and a covariance model. Rfam was originally developed 20 years ago to annotate genomes with ncRNAs using the covariance models. However, it has become the de-facto reference database for known ncRNAs and their alignments. This has led to it being used in new contexts including, RNA 3D structure prediction. This has pushed Rfam in new directions. Recently, Rfam has been improved by aligning sequences and base pair annotations from 3D structures into seed alignments. This connects Rfam alignments with 3D structures directly and allows improvements of families. We have used this to improve over 30 families and have started annotating pseudoknots. However despite these improvements, Rfam still has several limitations that make the prediction of RNA 3D structures challenging. Briefly, they are that ncRNA data is limited, biassed and incomplete. In this talk we will discuss some of these issues, suggest possible improvements, and challenge the community to solve them.

RNA molecules, essential players in the intricate machinery of cellular processes, exhibit a remarkable level of complexity in their three-dimensional structures. For many years, the primary focus in RNA structure study has traditionally been on base-pairing interactions and simple structure motifs. However, recent advances have unveiled another dimension of complexity – the presence of entanglements within RNA 3D structures. These structural intricacies, reminiscent of topological puzzles, may have profound implications for RNA function and dynamics. On the other hand, some of their types may be bugs injected into the structure, during its determination or in silico modeling process.

In this presentation, we will explore the diverse range of entangled motifs that can be found within RNA molecules. We will delve into the computational algorithms that have been developed to detect and analyze these unusual topological configurations in RNA structures. Finally, we will take a look at entanglements in experimental and simulated models of RNA 3D structure and we will learn if they can be untangled with any existing methods.

The physicochemical information of RNA molecules is greatly enhanced by posttranscriptional modifications, contributing to explain the diversity of their structures and functions.

To date, over 150 natural modifications have been characterized in all major classes of RNAs, ranging from isomerization or methylation, to the addition of bulky and complex chemical groups [1-2]. Modifications can change the folding landscape of RNA, resulting at times in alternative conformations [2-4]. This occurs by altering the interactions between nucleotides. Especially the H-bonding between nucleobases, both the regular Watson–Crick pairs enclosed in the RNA stems and the non Watson–Crick pairs outside the stems [5] - also known as tertiary interactions -, can be affected by modifications due to steric and energetic effects.

In order to investigate the impact of modifications on the interbase H-bonding, we have set up an approach combining structural bioinformatics with quantum mechanics (QM) calculations. Specifically, occurrences and structural context of modified base pairs (MBPs), i.e. base pairs featuring posttranscriptional modifications, are collected from the RNA structures in the PDB and classified by bioinformatics tools. Then, QM calculations are performed to clarify the effect of the modification on the geometry and stability of the corresponding base pair. We have applied this approach over time to both natural and non-natural (synthetic) modifications (see for instance [6-7]) and, in 2015, we have presented an atlas of MBPs, i.e. a systematic study of all the MBPs in RNA experimental structures [8]. At the time, we could identify a total of »900 occurrences for 11 natural modifications, with roughly half of them involved in base pairing. Our atlas 1.0 consisted of 27 MBPs, unique in terms of identity of H-bonded bases and/or geometry classification.

Herein, to extend our understanding of how posttranscriptional modifications act on the structure of RNA molecules to influence their function, we present an updated atlas, derived from an over doubled structures dataset. It consists overall of almost 100 unique MBPs, featuring 35 different posttranscriptional modifications, located in a variety of different RNA molecules and structural motifs. Consistently with our previous findings, most of the MBPs are non Watson–Crick like and are involved in RNA tertiary structure motifs. Results of the structural analyses, along with insight from QM calculations into the impact of the different modifications on the geometry and stability of the corresponding base pairs, will be presented and discussed.

1. P. Boccaletto, F. Stefaniak, A. Ray, A. Cappannini, S. Mukherjee, et al., Nucleic Acids Res., 50, (2022), D231.

2. P. F. Agris, RNA, 21, (2015), 552.

3. M. Helm, Nucleic Acids Res., 34, (2006), 721.

4. K. I. Zhou, M. Parisien, Q. Dai, N. Liu, L. Diatchenko, J. R. Sachleben, T. Pan, J. Mol. Biol, 428, (2016), 822.

5. N. B. Leontis, J. Stombaugh, E. Westhof, Nucleic Acids Res., 30, (2002), 3497.

6. R. Oliva, L. Cavallo, A. Tramontano, Nucleic Acids Res., 34, (2006), 865.

7. M. Chawla, S. Gorle, A. R. Shaikh, R. Oliva, L. Cavallo, Comput. Struct. Biotechnol. J., 19, (2021), 1312.

8. M. Chawla, R. Oliva, J. M. Bujnicki, L. Cavallo, Nucleic Acids Res., 43, (2015), 6714.

RNA adopts three-dimensional structures that play a crucial and direct role in its biological function. Understanding these diverse functions is necessary for the development of RNA-based therapies, but the complex structure of RNA molecules remains a major challenge. Computational methods have been developed throughout the years to fill the gap between the huge amount of known RNA sequences and their structures. With the increased number of RNA structures that are still to be discovered, predictive methods need to be robust and to be able to generalize to unseen new RNA families.

While structure predictions are a vast and complex problem, the evaluation and assessment of structure nativity is also at stake. RNA structure is a 3D object where the evaluation of a prediction has been discussed for years. Current methods rely on the comparison of a reference solved structure with a prediction, categorised as metrics. It can compare deviation on atoms like RMSD or εRMSD [1], or overlaps between them like CLASH score [2]. Other metrics are inspired by protein 3D evaluation metrics from the CASP competition. Indeed, RNA and protein 3D structures share common properties as 3D objects and adaptation of the known protein’s metrics like TM-score [3] can be done to RNA. It remains structural differences between protein and RNA molecules that hamper the full efficiency of structural evaluation metrics. RNA-oriented metrics have been developed to take full advantage of structural specificities like INF [4] or MCQ [5] scores.

Nonetheless, the metrics rely on a known solved structure, which in practice is not available. Predictive models are also based on the generation of multiple structures before selecting the best ones. Common approaches are thus to replicate the molecule free energy, where a minimum of energy would mean a stabilisation in the structure. This adaptation of the free energy of the structure has become a standard in the ranking, filtering and confidence assessment of structures. It often uses knowledge-based statistical potentials, with the requirement of a reference state to simulate structures without native interactions. This is the case for NAST [6], 3dRNAScore [7], DFIRE-RNA [8] and rsRNASP [9]. Recent advances tend to use deep learning to prevent manual pre-processing of RNA features like RNA3DCNN [10] or ARES [11].

RNA 3D structures remain of high complexity, and there is not a single existing metric or energy that could evaluate correctly all the available structures. Metrics and energies can be redundant between each other, while also complementary for structure assessment. The different existing metrics can be required to develop and understand predictive models’ weaknesses, while the diverse energies could help improve models’ generation such as the filtering process.

The current metrics and energies are the results of years of research by various groups. Each work has been developed in different programming languages, with different installation procedures and library versions that have evolved over the years. The installation process can be laborious for the community and is multiplied by the number of different metrics and energies. Efforts should be made on developing predictive models while engineering aspects for structures assessment should not be a bottleneck. Works have been done by the community with the development of RNAPuzzles [12], a CASP-like competition for RNA 3D structure assessment. It comes with RNA-tools [13], a centralised platform that tries to include the available RNA 3D structure related works. Nonetheless, it is limited in practice with the need to manually include binary files; that depend on the operating system of the user. There are also web servers available for some metrics and energies, which are useful for non-coder users. However, it limits the automation procedure, which should be considered due to the increasing number of solved 3D structures.

To help the development and the automation of RNA 3D structures evaluation, we have developed RNAdvisor: a software usable with one command line and that can compute both metrics and energies for given RNA. It uses eight existing codes written in C++, Java or Python and gathers them into a single interface. All the laborious installations are done in different stages of the Dockerfile. It leverages Docker containers for easy installation across diverse operating systems, simplifying accessibility for all researchers. It enables researchers to access both metrics and energies in one line of code, with customizable parameters to suit individual preferences.

RNAdvisor represents a significant advancement for the automation of RNA 3D structure evaluation. It offers a unified tool that enhances the accessibility of existing metrics and energies. It helps accelerate investigation in RNA 3D structure predictions.

The source code is available at: https://github.com/EvryRNA/rnadvisor.

1. Bottaro S, Di Palma F, and Bussi G. The Role of Nucleobase Interactions in RNA Structure and Dynamics. Nucleic acids research 2014;42.2. Davis IW, Leaver-Fay A, Chen VB, et al. MolProbity: all-atom contacts and structure validation for proteins and nucleic acids. Nucleic Acids Research 2007;35:W375–W383.3. Zhang Y and Skolnick J. Scoring function for automated assessment of protein structure template quality. Proteins 2004;57:702–10.

4. Parisien M, Cruz J, Westhof E, and Major F. New metrics for comparing and assessing discrepancies between RNA 3D structures and models. RNA (New York, N.Y.) 2009;15:1875–85

5. Zok T, Popenda M, and Szachniuk M. MCQ4Structures to compute similarity of molecule structures. Central European Journal of Operations Research 2013;22.

6. Jonikas MA, Radmer RJ, Laederach A, et al. Coarse-grained modeling of large RNA molecules with knowledge-based potentials and structural filters. RNA 2 2009;15:189–99.

7. Wang J, Zhao Y, Zhu C, and Xiao Y. 3dRNAscore: a distance and torsion angle dependent evaluation function of 3D RNA structures. Nucleic Acids Research 10 2015;43:e63–e63

8. Capriotti E, Norambuena T, Marti-Renom MA, and Melo F. All-atom knowledge-based potential for RNA structure prediction and assessment. Bioinformatics 2011; 27:1086–93.

9. Tan YL, Wang X, Shi YZ, Zhang W, and Tan ZJ. rsRNASP: A residue-separation based statistical potential for RNA 3D structure evaluation. Biophysical Journal 1 2022;121:142–56.

10. Li J, Zhu W, Wang J, et al. RNA3DCNN: Local and global quality assessments of RNA 3D structures using 3D deep convolutional neural networks. PLOS Computational Biology 2018;14:1–18.

11. Townshend RJL, Eismann S, Watkins AM, et al. Geometric deep learning of RNA structure. Science 6558 2021;373:1047–51.

12. Cruz J, Blanchet MF, Boniecki M, et al. RNA-Puzzles: A CASP-like evaluation of RNA three-dimensional structure prediction. RNA (New York, N.Y.) 2012;18:610–25.

13. Magnus M. rna-tools.online: a Swiss army knife for RNA 3D structure modeling workflow. Nucleic Acids Research 2022;50:W657–W662.

Accurately predicting the secondary structure of RNA, particularly for long non-coding RNA, has direct implications in healthcare, where it can be used for diagnostic, therapeutic, and drug discovery purposes. However, the majority of previous approaches are too costly in terms of computation budget to cope with the increasing complexity of long RNAs, and the ones that can scale to long RNAs lack accuracy to reliably predict their structures. We propose a new approach combining recursive cutting and machine learning techniques for predicting the secondary structures of long non-coding RNAs. In comparison, our method proves to be computationally efficient by recursively partitioning a sequence into smaller fragments until they can be easily managed by an existing model. We perform a benchmark of different state-of-the-art models and show that our approach indeed demonstrates better performance for long RNAs and a potential to bring significant improvements in the future, as well as interesting enhancing properties, which we discuss.