StructRNAfinder: an automated pipeline and web server for RNA families prediction

Noncoding RNAs (ncRNAs) are present in all domains of life, playing a critical role in the fine-tuning regulation of biological processes [1]. Their mode of action varies according to the RNA family it belongs. In 2005, the Rfam database created a limited type ontology to better represent the thousand of families identified so far and stored in the database [2]. Briefly, non-coding RNA genes (Gene) are composed by bona-fide RNAs with a recognised function (e.g. CRISPR, miRNAs, ribozymes, rRNAs, snoRNAs); structured cis-regulatory elements (Cis-reg), are represented by structural regulatory motifs available in RNA sequences (e.g. frameshift elements, riboswitches, thermoregulators); and Intron, composed by self-splicing RNAs. The prediction of RNA families in genome or transcriptome sequences often depends on its primary sequence conservation or secondary structural motifs. Thus, several bioinformatics workflows that use third-party software were created to predict or annotate different RNA classes using sequence or structure comparisons [2–5].

Secondary structure-based methods are critical for the annotation of specific regulatory RNAs [6, 7]. These approaches employ nucleotide sequences folding and minimum free energy calculation, in order to predict most of well-known RNA families [8]. For instance, microRNAs (miRNAs) have a characteristic structure of highly paired nucleotides, forming a double strand RNA structure from a single molecule of approximately 100 nt. Further, snoRNAs (small nucleolar RNAs) present a big loop associated to their binding with ribosomal RNAs (rRNAs). These structural features are of key importance for their ab-initio prediction and classification [9–11], sometimes coupled with other features, such as nucleotides conservation and covariation [12, 13]. Tools such as Infernal [14] utilize nucleotide sequences and/or secondary structure covariance models from known RNA families, like those available in Rfam and other databases [15, 16]. These tools directly compare known RNA families and nucleotide sequences, resulting in the identification of potential novel regulatory RNAs in genome or transcriptome sequences. However, large-scale prediction and annotation of regulatory noncoding RNAs can represent a daunting task, as multiple tools are required along the process. In addition, the constant parsing and processing of intermediate files needed to run these tools impose a great obstacle for researchers with lesser computational or bioinformatics backgrounds.

To avoid these technical bottlenecks, we developed an automated pipeline named StructRNAfinder. The tool was implemented using Perl scripts and third-party software and is focused on the identification and complete annotation of regulatory RNA families from transcriptome or genome sequences. We also implemented a user-friendly web server that allows users with no bioinformatics skills to perform all analyses. Both stand-alone and web server StructRNAfinder versions can be accessed at http://structrnafinder.integrativebioinformatics.me.

In this work, we described a new automated pipeline, named StructRNAfinder, which was developed to facilitate the identification and complete annotation of regulatory RNA families available in nucleotide sequences (DNA or RNA). When predicting/annotating RNAs in nucleotides sequences, the user needs to manipulate several plain text results generated by different tools used on the process, which are difficult to visualize and manipulate in typical text viewers rather than Linux command line. The advantage of StructRNAfinder relies on processing and integrating the data obtained by each tool and database employed along the workflow, of which several files are generated and displayed in user-friendly reports, useful for downstream analyses and data exploration.

We have successfully applied the stand-alone version of this automated pipeline on analyses of the noncoding RNA content in genomic sequences from the most diverse set of organisms, covering all three domains of life, plus viruses (unpublished and published data). For instance, on the development of LeishDB [20], a reference database for Leishmania braziliensis genomic information; we used this tool, together with other strategies, to obtain the most comprehensive characterization of noncoding RNAs for Leishmania species. Analyses using genomic and transcriptomic E. coli datasets allowed the prediction of 488 different RNAs (cmsearch e-value cutoff of 0.01), part of 184 different Rfam families, on the genomic sequence of E. coli str k-12; and the correct prediction of 98.05% of experimentally validated RNA transcripts (i.e. predicted with the same functional annotation as provided by the authors [19]). The remaining 2% are related to validated RNAs in which their covariance models are not yet made available in Rfam database.

StructRNAfinder is presented in both stand-alone and web server versions, facilitating the usage for all kind of users. The web-based version allows the user-friendly prediction of RNA families available in sequences of up to 10,000,000 nucleotides (10 Mb), which is enough to predict the repertoire of regulatory RNAs in the sequenced genomes of all Archaea and most of Bacteria available in NCBI. Indeed, Rfam allows a batch sequence search on their web server. However, it does not allow the usage of different filtering options available in StructRNAfinder (i.e. Infernal search method, cutoff filters, strand-specific search, best hit per sequence selection), and its search is limited to 200,000 nucleotides (200 Kb). A table comparing webStructRNAfinder, structRNAfinder and Rfam batch search is available in Additional file 4.

The advantages of StrucRNAfinder stand-alone compared to its web-based version are the possibility to analyse large genomes, e.g. eukaryotic organisms; to perform large-scale analyses using several genomes; and to use as input the covariance models generated by the user itself, instead of those generated by Rfam. Indeed, recently, Eggenhofer and collaborators developed an unsupervised tool that allows the generation of covariance models using a single sequence as input [21]. It collects potential RNA family members from a model generated based on multiple interactions of homology searches against nucleotide sequences of taxonomy related organisms available in NCBI, and structural alignments analyses, in order to generate the most suitable covariance model for that particular RNA sequence. For instance, this is useful for structural conservation analysis of new ncRNAs found in transcriptomes, in which users can easily generate a covariance model of one or a set of transcripts of interest and search for its presence in different sets of genome sequences using the stand-alone version of StructRNAfinder.

StructRNAfinder facilitates de prediction and complete functional annotation of RNA families in nucleotides sequences, by integrating different tools and databases commonly used on the prediction and functional annotation of regulatory RNAs. One main advantage over other existing tools relies on the large-scale processing and integration of the data obtained by each tool and database employed along the workflow, with useful files generated and displayed in user-friendly reports helpful for downstream analyses. The automatic generations of these reports avoid the time-consuming process of writing scripts for parsing the output and input files for the tools and databases employed, especially for users without any programming or bioinformatics skills. These features facilitate the genome-wide predictions of the complete repertoire of RNA families available in small and large genomes and also in assembled transcriptomes derived from NGS studies.

Project name: StructRNAfinder.

Project home page: http://structrnafinder.integrativebioinformatics.me

Operating system(s): LINUX.

Programming language: Perl, HTML5, PHP5 and JavaScript.

Other requirements: Linux Perl package libgd-perl version 2.53, Perl package Bio::Graphics 2.4, Infernal tool version 1.1, Vienna package version 2.1.8, Rfam covariance models version 12.

License: GNU GPL version 3.