Prediction of lncRNAs and their interactions with nucleic acids: benchmarking bioinformatics tools. (24th April 2018)

• Record Type: Journal Article
• Authors: Antonov, Ivan V; Mazurov, Evgeny; Borodovsky, Mark; Medvedeva, Yulia A
• Is Part Of: Briefings in bioinformatics. Volume 20:Number 2(2019)
• Page Start: 551
• Page End: 564
• Publication Date: 2018-04-24
• Subjects: lncRNA -- RNA-RNA interaction -- RNA-DNA interaction -- gene prediction; Genetics -- Data processing -- Periodicals; Molecular biology -- Data processing -- Periodicals; Genomes -- Data processing -- Periodicals; 572.80285
• DOI: 10.1093/bib/bby032
• Languages: English
• ISSNs: 1467-5463

Abstract

Abstract: The genomes of mammalian species are pervasively transcribed producing as many noncoding as protein-coding RNAs. There is a growing body of evidence supporting their functional role. Long noncoding RNA (lncRNA) can bind both nucleic acids and proteins through several mechanisms. A reliable computational prediction of the most probable mechanism of lncRNA interaction can facilitate experimental validation of its function. In this study, we benchmarked computational tools capable to discriminate lncRNA from mRNA and predict lncRNA interactions with other nucleic acids. We assessed the performance of 9 tools for distinguishing protein-coding from noncoding RNAs, as well as 19 tools for prediction of RNA-RNA and RNA-DNA interactions. Our conclusions about the considered tools were based on their performances on the entire genome/transcriptome level, as it is the most common task nowadays. We found that FEELnc and CPAT distinguish between coding and noncoding mammalian transcripts in the most accurate manner. ASSA, RIBlast and LASTAL, as well as Triplexator, turned out to be the best predictors of RNA-RNA and RNA-DNA interactions, respectively. We showed that the normalization of the predicted interaction strength to the transcript length and GC content may improve the accuracy of inferring RNA interactions. Yet, all the current tools have difficulties to make accurate predictions of short- trans RNA-RNA interactions—stretches of sparse contacts. All over, there is still room for improvement in each category, especially for predictions of RNA interactions.