Accurate identification of RNA editing sites from primitive sequence with deep neural networks

Harnessing ADAR-Mediated Site-Specific RNA Editing in Immune-Related Disease: Prediction and Therapeutic Implications

ADAR (Adenosine Deaminases Acting on RNA) proteins are a group of enzymes that play a vital role in RNA editing by converting adenosine to inosine in RNAs. This process is a frequent post-transcriptional event observed in metazoan transcripts. Recent studies indicate widespread dysregulation of ADAR-mediated RNA editing across many immune-related diseases, such as human cancer. We comprehensively review ADARs' function as pattern recognizers and their capability to contribute to mediating immune-related pathways. We also highlight the potential role of site-specific RNA editing in maintaining homeostasis and its relationship to various diseases, such as human cancers. More importantly, we summarize the latest cutting-edge computational approaches and data resources for predicting and analyzing RNA editing sites. Lastly, we cover the recent advancement in site-directed ADAR editing tool development. This review presents an up-to-date overview of ADAR-mediated RNA editing, how site-specific RNA editing could potentially impact disease pathology, and how they could be harnessed for therapeutic applications.

Benchmarking RNA Editing Detection Tools

RNA, like DNA and proteins, can undergo modifications. To date, over 170 RNA modifications have been identified, leading to the emergence of a new research area known as epitranscriptomics. RNA editing is the most frequent RNA modification in mammalian transcriptomes, and two types have been identified: (1) the most frequent, adenosine to inosine (A-to-I); and (2) the less frequent, cysteine to uracil (C-to-U) RNA editing. Unlike other epitranscriptomic marks, RNA editing can be readily detected from RNA sequencing (RNA-seq) data without any chemical conversions of RNA before sequencing library preparation. Furthermore, analyzing RNA editing patterns from transcriptomic data provides an additional layer of information about the epitranscriptome. As the significance of epitranscriptomics, particularly RNA editing, gains recognition in various fields of biology and medicine, there is a growing interest in detecting RNA editing sites (RES) by analyzing RNA-seq data. To cope with this increased interest, several bioinformatic tools are available. However, each tool has its advantages and disadvantages, which makes the choice of the most appropriate tool for bench scientists and clinicians difficult. Here, we have benchmarked bioinformatic tools to detect RES from RNA-seq data. We provide a comprehensive view of each tool and its performance using previously published RNA-seq data to suggest recommendations on the most appropriate for utilization in future studies.

ATTIC is an integrated approach for predicting A-to-I RNA editing sites in three species

A-to-I editing is the most prevalent RNA editing event, which refers to the change of adenosine (A) bases to inosine (I) bases in double-stranded RNAs. Several studies have revealed that A-to-I editing can regulate cellular processes and is associated with various human diseases. Therefore, accurate identification of A-to-I editing sites is crucial for understanding RNA-level (i.e. transcriptional) modifications and their potential roles in molecular functions. To date, various computational approaches for A-to-I editing site identification have been developed; however, their performance is still unsatisfactory and needs further improvement. In this study, we developed a novel stacked-ensemble learning model, ATTIC (A-To-I ediTing predICtor), to accurately identify A-to-I editing sites across three species, including Homo sapiens, Mus musculus and Drosophila melanogaster. We first comprehensively evaluated 37 RNA sequence-derived features combined with 14 popular machine learning algorithms. Then, we selected the optimal base models to build a series of stacked ensemble models. The final ATTIC framework was developed based on the optimal models improved by the feature selection strategy for specific species. Extensive cross-validation and independent tests illustrate that ATTIC outperforms state-of-the-art tools for predicting A-to-I editing sites. We also developed a web server for ATTIC, which is publicly available at http://web.unimelb-bioinfortools.cloud.edu.au/ATTIC/. We anticipate that ATTIC can be utilized as a useful tool to accelerate the identification of A-to-I RNA editing events and help characterize their roles in post-transcriptional regulation.

EditPredict: Prediction of RNA editable sites with convolutional neural network

RNA editing exerts critical impacts on numerous biological processes. While millions of RNA editings have been identified in humans, much more are expected to be discovered. In this work, we constructed Convolutional Neural Network (CNN) models to predict human RNA editing events in both Alu regions and non-Alu regions. With a validation dataset resulting from CRISPR/Cas9 knockout of the ADAR1 enzyme, the validation accuracies reached 99.5% and 93.6% for Alu and non-Alu regions, respectively. We ported our CNN models in a web service named EditPredict. EditPredict not only works on reference genome sequences but can also take into consideration single nucleotide variants in personal genomes. In addition to the human genome, EditPredict tackles other model organisms including bumblebee, fruitfly, mouse, and squid genomes. EditPredict can be used stand-alone to predict novel RNA editing and it can be used to assist in filtering for candidate RNA editing detected from RNA-Seq data.

RDDSVM: accurate prediction of A-to-I RNA editing sites from sequence using support vector machines

Adenosine to inosine (A-to-I) editing in RNA is involved in various biological processes like gene expression, alternative splicing, and mRNA degradation associated with carcinogenesis and various human diseases. Therefore, accurate identification of RNA editing sites in transcriptome is valuable for research and medicine. RNA-seq is very useful for the detection of RNA editing events in condition-specific cells. However, computational analysis methods of RNA-seq data have considerable false-positive risks due to mapping errors. In this study, we developed a simple machine learning method using support vector machines to train sequence and structure information derived from flanking sequences of experimentally verified A-to-I editing sites to predict new A-to-I editing sites in RNA. The highest performance results were obtained by the model that utilizes the composition of the triplet sequence elements in the flanking regions of the in A-to-I editing sites. Using this model, the SVM classifier also showed high performance on experimentally verified data providing a sensitivity of 92.8%, specificity of 77.1%, and accuracy of 90.2%. To compare the predictive capacity of our method with other classifiers that use sequence information, we have used validated human A-to-I RNA editing sites by Sanger sequencing. Out of 58 validated editing sites, our method recognized 53 of them correctly with an accuracy of 91.4% outperforming other classifiers. As to our knowledge, this is the first case of utilization of the composition of the triplet sequence elements neighboring A-to-I editing sites for the prediction of new A-to-I editing sites in RNA. The methodology is very easy to perform and computationally low demanding making it a convenient and valuable choice for facilities with low sources. To facilitate the usage of the method publicly, we developed an open-source program called RDDSVM to perform prediction on candidate A-to-I RNA editing sites using support vector machines.

A-to-I RNA Editing in Cancer: From Evaluating the Editing Level to Exploring the Editing Effects

As an important regulatory mechanism at the posttranscriptional level in metazoans, adenosine deaminase acting on RNA (ADAR)-induced A-to-I RNA editing modification of double-stranded RNA has been widely detected and reported. Editing may lead to non-synonymous amino acid mutations, RNA secondary structure alterations, pre-mRNA processing changes, and microRNA-mRNA redirection, thereby affecting multiple cellular processes and functions. In recent years, researchers have successfully developed several bioinformatics software tools and pipelines to identify RNA editing sites. However, there are still no widely accepted editing site standards due to the variety of parallel optimization and RNA high-seq protocols and programs. It is also challenging to identify RNA editing by normal protocols in tumor samples due to the high DNA mutation rate. Numerous RNA editing sites have been reported to be located in non-coding regions and can affect the biosynthesis of ncRNAs, including miRNAs and circular RNAs. Predicting the function of RNA editing sites located in non-coding regions and ncRNAs is significantly difficult. In this review, we aim to provide a better understanding of bioinformatics strategies for human cancer A-to-I RNA editing identification and briefly discuss recent advances in related areas, such as the oncogenic and tumor suppressive effects of RNA editing.

The epitranscriptome beyond m6A

Following its transcription, RNA can be modified by >170 chemically distinct types of modifications - the epitranscriptome. In recent years, there have been substantial efforts to uncover and characterize the modifications present on mRNA, motivated by the potential of such modifications to regulate mRNA fate and by discoveries and advances in our understanding of N ⁶-methyladenosine (m⁶A). Here, we review our knowledge regarding the detection, distribution, abundance, biogenesis, functions and possible mechanisms of action of six of these modifications - pseudouridine (Ψ), 5-methylcytidine (m⁵C), N ¹-methyladenosine (m¹A), N ⁴-acetylcytidine (ac⁴C), ribose methylations (N_m) and N ⁷-methylguanosine (m⁷G). We discuss the technical and analytical aspects that have led to inconsistent conclusions and controversies regarding the abundance and distribution of some of these modifications. We further highlight shared commonalities and important ways in which these modifications differ with respect to m⁶A, based on which we speculate on their origin and their ability to acquire functions over evolutionary timescales.

Chemical Profiling of A-to-I RNA Editing Using a Click-Compatible Phenylacrylamide

Straightforward methods for detecting adenosine-to-inosine (A-to-I) RNA editing are key to a better understanding of its regulation, function, and connection with disease. We address this need by developing a novel reagent, N-(4-ethynylphenyl)acrylamide (EPhAA), and illustrating its ability to selectively label inosine in RNA. EPhAA is synthesized in a single step, reacts rapidly with inosine, and is "click"-compatible, enabling flexible attachment of fluorescent probes at editing sites. We first validate EPhAA reactivity and selectivity for inosine in both ribonucleosides and RNA substrates, and then apply our approach to directly monitor in vitro A-to-I RNA editing activity using recombinant ADAR enzymes. This method improves upon existing inosine chemical-labeling techniques and provides a cost-effective, rapid, and non-radioactive approach for detecting inosine formation in RNA. We envision this method will improve the study of A-to-I editing and enable better characterization of RNA modification patterns in different settings.

Investigating RNA editing in deep transcriptome datasets with REDItools and REDIportal

RNA editing is a widespread post-transcriptional mechanism able to modify transcripts through insertions/deletions or base substitutions. It is prominent in mammals, in which millions of adenosines are deaminated to inosines by members of the ADAR family of enzymes. A-to-I RNA editing has a plethora of biological functions, but its detection in large-scale transcriptome datasets is still an unsolved computational task. To this aim, we developed REDItools, the first software package devoted to the RNA editing profiling in RNA-sequencing (RNAseq) data. It has been successfully used in human transcriptomes, proving the tissue and cell type specificity of RNA editing as well as its pervasive nature. Outcomes from large-scale REDItools analyses on human RNAseq data have been collected in our specialized REDIportal database, containing more than 4.5 million events. Here we describe in detail two bioinformatic procedures based on our computational resources, REDItools and REDIportal. In the first procedure, we outline a workflow to detect RNA editing in the human cell line NA12878, for which transcriptome and whole genome data are available. In the second procedure, we show how to identify dysregulated editing at specific recoding sites in post-mortem brain samples of Huntington disease donors. On a 64-bit computer running Linux with ≥32 GB of random-access memory (RAM), both procedures should take ~76 h, using 4 to 24 cores. Our protocols have been designed to investigate RNA editing in different organisms with available transcriptomic and/or genomic reads. Scripts to complete both procedures and a docker image are available at https://github.com/BioinfoUNIBA/REDItools.

RNA editing in the forefront of epitranscriptomics and human health

Post-transcriptional modifications have been recently expanded with the addition of RNA editing, which is predominantly mediated by adenosine and cytidine deaminases acting on DNA and RNA. Here, we review the full spectrum of physiological processes in which these modifiers are implicated, among different organisms. Adenosine to inosine (A-to-I) editors, members of the ADAR and ADAT protein families are important regulators of alternative splicing and transcriptional control. On the other hand, cytidine to uridine (C-to-U) editors, members of the AID/APOBEC family, are heavily implicated in innate and adaptive immunity with important roles in antibody diversification and antiviral response. Physiologically, these enzymes are present in the nucleus and/or the cytoplasm, where they modify various RNA molecules, including miRNAs, tRNAs apart from mRNAs, whereas DNA editing is also possible by some of them. The expansion of next generation sequencing technologies provided a wealth of data regarding such modifications. RNA editing has been implicated in various disorders including cancer, and neurological diseases of the brain or the central nervous system. It is also related to cancer heterogeneity and the onset of carcinogenesis. Response to treatment can also be affected by the RNA editing status where drug efficacy is significantly compromised. Studying RNA editing events can pave the way to the identification of new disease biomarkers, and provide a more personalised therapy to various diseases.

Bioinformatic methods for cancer neoantigen prediction

Tumor cells accumulate aberrations not present in normal cells, leading to presentation of neoantigens on MHC molecules on their surface. These non-self neoantigens distinguish tumor cells from normal cells to the immune system and are thus targets for cancer immunotherapy. The rapid development of molecular profiling platforms, such as next-generation sequencing, has enabled the generation of large datasets characterizing tumor cells. The simultaneous development of algorithms has enabled rapid and accurate processing of these data. Bioinformatic software tools encoding the algorithms can be strung together in a workflow to identify neoantigens. Here, with a focus on high-throughput sequencing, we review state-of-the art bioinformatic tools along with the steps and challenges involved in neoantigen identification and recognition.

Computational approaches for detection and quantification of A-to-I RNA-editing

Adenosine deaminases that act on RNA (ADARs) catalyze adenosine-to-inosine (A-to-I) RNA editing in double-stranded RNA. Such editing is important for protection against false activation of the immune system, but also confers plasticity on the transcriptome by generating several versions of a transcript from a single genomic locus. Recently, great efforts were made in developing computational methods for detecting editing events directly from RNA-sequencing (RNA-seq) data. These efforts have led to an improved understanding of the makeup of the editome in various genomes. Here we review recent advances in editing detection based on the data available to the researcher, with emphasis on the principles underlying the various methods and the limitations they were designed to overcome. We also discuss the available various methods for analyzing and quantifying editing levels. This review collects and organizes the available approaches for analyzing RNA editing and discuss the current status of the different A-to-I detection methods with possible directions for extending these approaches.