MOTIVATION: Long-read RNA sequencing enables the mapping of RNA modifications, structures, and protein-interaction sites at the resolution of individual transcript isoforms. To understand the functions of these RNA features, it is critical to analyze them in the context of transcriptomic and genomic annotations, such as open reading frames and splice junctions. RESULTS: We have developed R2Dtool, a bioinformatics tool that integrates transcript-mapped information with transcript and genome annotations, allowing for the isoform-resolved analytics and graphical representation of RNA features in their genomic context. We illustrate R2Dtool's capability to integrate and expedite RNA feature analysis using epitranscriptomics data. R2Dtool facilitates the comprehensive analysis and interpretation of alternative transcript isoforms. AVAILABILITY AND IMPLEMENTATION: R2Dtool is freely available under the MIT license at github.com/comprna/R2Dtool.

The expanding field of epitranscriptomics might rival the epigenome in the diversity of biological processes impacted. In recent years, the development of new high-throughput experimental and computational techniques has been a key driving force in discovering the properties of RNA modifications. Machine learning applications, such as for classification, clustering or de novo identification, have been critical in these advances. Nonetheless, various challenges remain before the full potential of machine learning for epitranscriptomics can be leveraged. In this review, we provide a comprehensive survey of machine learning methods to detect RNA modifications using diverse input data sources. We describe strategies to train and test machine learning methods and to encode and interpret features that are relevant for epitranscriptomics. Finally, we identify some of the current challenges and open questions about RNA modification analysis, including the ambiguity in predicting RNA modifications in transcript isoforms or in single nucleotides, or the lack of complete ground truth sets to test RNA modifications. We believe this review will inspire and benefit the rapidly developing field of epitranscriptomics in addressing the current limitations through the effective use of machine learning.

The convergence of artificial intelligence (AI) and machine learning (ML) with the field of RNA biology has ushered in a new era of scientific discovery, providing unprecedented insights into the intricacies of RNA regulation at the molecular level. By analyzing vast datasets, AI/ML offers unprecedented opportunities to decode RNA’s complexities: characterize the composition and structure of RNA, uncover their function, decipher the intricate regulatory networks that govern their activity and determine their relevance and alterations in disease. As a result, RNA is emerging as a central player in understanding cellular function and providing versatile targets and vectors for novel therapeutics. This special issue provides in-depth reviews and assessments of AI/ML methodologies applied to various aspects of RNA biology, including the RNA chemical composition and structure, interactions with proteins and the individual and collective roles of RNA molecules in pathogenesis and disease outcomes. This collection highlights the biological relevance of studying RNA and covers essential aspects of AI/ML development in RNA biology and biomedical sciences in general. The latter comprises the choice of model types and training datasets, interpretability, the definition of the prediction tasks, the advantages and limitations of deep learning and classical approaches and the integration with biophysical properties and domain knowledge. One of the most effective applications of AI/ML in RNA biology is RNA structure prediction. The self-pairing of RNA bases and three-dimensional folding are key determinants of the molecular functions of RNAs, including their stability and interactions with protein, RNA and other molecules, and have implications for RNA drug discovery [1]. AI/ML algorithms can predict RNA structures more rapidly and accurately than traditional methods, enabling discoveries from large datasets. In ‘Recent trends in RNA informatics: A review of machine learning and deep learning for RNA secondary structure prediction and RNA drug discovery’, Sato and Hamada [2] review the use of ML and deep learning techniques for RNA secondary structure prediction and RNA drug discovery, with a focus on the de novo prediction of secondary structures in individual RNA molecules. In ‘Machine learning modeling of RNA structures: Methods, challenges, and future perspectives’, Wu et al. [3] review ML-based methods for predicting RNA secondary and tertiary structures, dissecting the critical elements for accurate prediction, highlighting the shortcomings of current design decisions, and proposing future directions for more accurate predictions. The authors also discuss integrating classical thermodynamic principles into the data-driven approaches of deep learning to improve model accuracy and generalization and bring up the need to address the lengthy runtimes required for long RNA sequences. Among the structures that RNA can adopt, highly stable four-stranded structures known as RNA G-quadruplexes (G4s) are possibly the least well-characterized. They play essential roles in various biological processes, including gene expression regulation, DNA replication, recombination and repair [4]. The recent development of transcriptome-wide experimental approaches to measure G4 formation has enabled the training of AI/ML models to characterize these structures and discover new ones. In ‘An overview on nucleic-acid G-quadruplex prediction: from rule-based methods to deep neural networks’, Elimelech-Zohar et al. [5] review the success of deep neural network applications for predicting DNA and RNA G4s and perform a comparison with previous methods based on rules. The authors also discuss the experimental technologies that generate comprehensive G4 datasets, which are essential to training deep neural networks, and discuss the use of interpretability techniques to extract relevant features and the potential of transfer learning from DNA G4s to RNA G4s. The chemical modifications of messenger RNAs (mRNAs) can affect splicing, stability and translation [6]. Describing the landscape of RNA modifications can provide insights into disease mechanisms, critical cellular processes and organism development. In ‘Concepts and methods for transcriptome-wide prediction of chemical messenger RNA modifications with Machine Learning’, Acera Mateos et al. [7] discuss different AI/ML strategies for predicting mRNA modifications. This review categorizes experiment-independent and experiment-based methods, discusses their advantages and applicability and describes the current open challenges in the field, such as interpretability and isoform-specific detection, and how AI/ML can address them. RNA fate and function are intimately coupled to interactions with protein complexes [8]. Identifying and characterizing the complex cellular networks of protein–RNA interactions is thus essential to describe critical biological processes, such as mRNA translation and localization, and their involvement in human disease. In ‘A systematic benchmark of AI/ML methods for protein–RNA interaction prediction’, Horlacher et al. [9] comprehensively evaluate 37 ML methods for predicting protein–RNA interactions using standardized datasets and evaluation metrics. This study provides researchers with rich information to help them choose the most suitable prediction method and assist developers in creating and improving AI/ML strategies for this task. The authors also describe how modeling the experimental biases appears to provide an advantage. At the same time, the accuracy of the predictive models varies according to the coverage, quality and representativeness of different experimental datasets used as inputs. Overall, the study provides insights into the impact of various factors on the performance of AI/ML methods for protein–RNA interaction prediction, guiding the development of future approaches in this field. Mutations in DNA may impact mRNA splicing, leading to aberrant splicing, functional alterations and disease [10]. Identification of mutations that alter splicing and prediction of their functional impact has enabled the interpretation of mutations of unknown function and improved the diagnostic rates in rare diseases. In ‘Splicing defects in rare diseases: transcriptomics and machine learning strategies toward genetic diagnosis’, Wang et al. [11] discuss the relevance of detecting and interpreting pathogenic splicing variants in rare genetic diseases and describe how the integration of RNA sequencing technologies and AI/ML strategies have provided fundamental advances in the field. The review highlights the need for further improvements in sequencing technologies and predictive modeling to enhance our understanding of splicing regulation and bridge the diagnostic gap for rare disease patients. Cellular and organismal phenotypes emerge from the intricate interplay and synergistic behaviors of thousands of RNA molecules in cells and tissues. AI/ML modeling allows one to study biological systems beyond the individual molecules to interpret their collective impacts and explain emergent properties and responses at the systems level. In ‘Benchmarking of analytical combinations for COVID-19 outcome prediction using single-cell RNA sequencing data’, Cao et al. [12] compare multiple AI/ML strategies to predict COVID-19 outcomes using single-cell RNA sequencing data. The authors evaluate the performance of various models, compare single-view and multi-view feature spaces and assess the impact of integration strategies when using multiple input datasets. The study aims to better understand disease mechanisms at an individual rather than a cell level, using methods that extract information from multiple samples and conditions. In conclusion, integrating AI/ML methodologies with RNA biology has opened exciting new avenues for exploration, advancing our understanding of the fundamental processes of life. This special issue provides comprehensive reviews and evaluations of AI/ML methodologies in RNA biology. Topics covered include the characterization of the RNA molecular composition, structure, and function, deciphering the protein networks governing RNA activity, and determining the roles of RNA in disease. Besides highlighting the fundamental roles of RNA, this article collection covers essential aspects of AI/ML development, including model choice, training and testing strategies, interpretability, and integration with complementary knowledge. The application of AI/ML methodologies in RNA biology has greatly improved our understanding of fundamental life processes and underscored the relevance of RNA in biomedical research and therapeutic development. We acknowledge support from the Australian Research Council (ARC) Discovery Project grants DP220101352 and DP210102385 (to EE), from the National Health and Medical Research Council (NHMRC) Ideas Grant 2018833 (to EE), and from the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) SPP 1935 ZA 881/5-2 (Projektnummer 273941853, to KZ).

Immunotherapies provide effective treatments for previously untreatable tumors and identifying tumor-specific epitopes can help elucidate the molecular determinants of therapy response. Here, we describe a pipeline, ISOTOPE (ISOform-guided prediction of epiTOPEs In Cancer), for the comprehensive identification of tumor-specific splicing-derived epitopes. Using RNA sequencing and mass spectrometry for MHC-I associated proteins, ISOTOPE identified neoepitopes from tumor-specific splicing events that are potentially presented by MHC-I complexes. Analysis of multiple samples indicates that splicing alterations may affect the production of self-epitopes and generate more candidate neoepitopes than somatic mutations. Although there was no difference in the number of splicing-derived neoepitopes between responders and non-responders to immune therapy, higher MHC-I binding affinity was associated with a positive response. Our analyses highlight the diversity of the immunogenic impacts of tumor-specific splicing alterations and the importance of studying splicing alterations to fully characterize tumors in the context of immunotherapies. ISOTOPE is available at https://github.com/comprna/ISOTOPE.

RNA-protein binding is critical to gene regulation, controlling fundamental processes including splicing, translation, localization and stability, and aberrant RNA-protein interactions are known to play a role in a wide variety of diseases. However, molecular understanding of RNA-protein interactions remains limited; in particular, identification of RNA motifs that bind proteins has long been challenging, especially when such motifs depend on both sequence and structure. Moreover, although RNA binding proteins (RBPs) often contain more than one binding domain, algorithms capable of identifying more than one binding motif simultaneously have not been developed. In this paper we present a novel pipeline to determine binding peaks in crosslinking immunoprecipitation (CLIP) data, to discover multiple possible RNA sequence/structure motifs among them, and to experimentally validate such motifs. At the core is a new semi-automatic algorithm SARNAclust, the first unsupervised method to identify and deconvolve multiple sequence/structure motifs simultaneously. SARNAclust computes similarity between sequence/structure objects using a graph kernel, providing the ability to isolate the impact of specific features through the bulge graph formalism. Application of SARNAclust to synthetic data shows its capability of clustering 5 motifs at once with a V-measure value of over 0.95, while GraphClust achieves only a V-measure of 0.083 and RNAcontext cannot detect any of the motifs. When applied to existing eCLIP sets, SARNAclust finds known motifs for SLBP and HNRNPC and novel motifs for several other RBPs such as AGGF1, AKAP8L and ILF3. We demonstrate an experimental validation protocol, a targeted Bind-n-Seq-like high-throughput sequencing approach that relies on RNA inverse folding for oligo pool design, that can validate the components within the SLBP motif. Finally, we use this protocol to experimentally interrogate the SARNAclust motif predictions for protein ILF3. Our results support a newly identified partially double-stranded UUUUUGAGA motif similar to that known for the splicing factor HNRNPC.

MOTIVATION: High-throughput sequencing (HTS) has revolutionized gene regulation studies and is now fundamental for the detection of protein-DNA and protein-RNA binding, as well as for measuring RNA expression. With increasing variety and sequencing depth of HTS datasets, the need for more flexible and memory-efficient tools to analyse them is growing. RESULTS: We describe Pyicos, a powerful toolkit for the analysis of mapped reads from diverse HTS experiments: ChIP-Seq, either punctuated or broad signals, CLIP-Seq and RNA-Seq. We prove the effectiveness of Pyicos to select for significant signals and show that its accuracy is comparable and sometimes superior to that of methods specifically designed for each particular type of experiment. Pyicos facilitates the analysis of a variety of HTS datatypes through its flexibility and memory efficiency, providing a useful framework for data integration into models of regulatory genomics. AVAILABILITY: Open-source software, with tutorials and protocol files, is available at http://regulatorygenomics.upf.edu/pyicos or as a Galaxy server at http://regulatorygenomics.upf.edu/galaxy CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The branch point (BP) is one of the three obligatory signals required for pre-mRNA splicing. In mammals, the degeneracy of the motif combined with the lack of a large set of experimentally verified BPs complicates the task of modeling it in silico, and therefore of predicting the location of natural BPs. Consequently, BPs have been disregarded in a considerable fraction of the genome-wide studies on the regulation of splicing in mammals. We present a new computational approach for mammalian BP prediction. Using sequence conservation and positional bias we obtained a set of motifs with good agreement with U2 snRNA binding stability. Using a Support Vector Machine algorithm, we created a model complemented with polypyrimidine tract features, which considerably improves the prediction accuracy over previously published methods. Applying our algorithm to human introns, we show that BP position is highly dependent on the presence of AG dinucleotides in the 3' end of introns, with distance to the 3' splice site and BP strength strongly correlating with alternative splicing. Furthermore, experimental BP mapping for five exons preceded by long AG-dinucleotide exclusion zones revealed that, for a given intron, more than one BP can be chosen throughout the course of splicing. Finally, the comparison between exons of different evolutionary ages and pseudo exons suggests a key role of the BP in the pathway of exon creation in human. Our computational and experimental analyses suggest that BP recognition is more flexible than previously assumed, and it appears highly dependent on the presence of downstream polypyrimidine tracts. The reported association between BP features and the splicing outcome suggests that this, so far disregarded but yet crucial, element buries information that can complement current acceptor site models.

BACKGROUND: Despite the continuous production of genome sequence for a number of organisms, reliable, comprehensive, and cost effective gene prediction remains problematic. This is particularly true for genomes for which there is not a large collection of known gene sequences, such as the recently published chicken genome. We used the chicken sequence to test comparative and homology-based gene-finding methods followed by experimental validation as an effective genome annotation method. RESULTS: We performed experimental evaluation by RT-PCR of three different computational gene finders, Ensembl, SGP2 and TWINSCAN, applied to the chicken genome. A Venn diagram was computed and each component of it was evaluated. The results showed that de novo comparative methods can identify up to about 700 chicken genes with no previous evidence of expression, and can correctly extend about 40% of homology-based predictions at the 5' end. CONCLUSIONS: De novo comparative gene prediction followed by experimental verification is effective at enhancing the annotation of the newly sequenced genomes provided by standard homology-based methods.