EcircPred: Sequence and secondary structural property based computational identification of exonic circular RNAs. (December 2018)

• Record Type: Journal Article
• Authors: Kumar, Rajnish; Lahiri, Tapobrata
• Is Part Of: Computational biology and chemistry. Volume 77(2018)
• Page Start: 28
• Page End: 35
• Publication Date: 2018-12
• Subjects: Exonic circular RNA prediction -- RNA secondary structure feature -- Di-nucleotide properties -- Elman recurrent backpropagation network -- Entropy of RNA -- Emission probability; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2018.08.002
• Languages: English
• ISSNs: 1476-9271

Abstract

Graphical abstract: Highlights: First machine learning model was introduced to discriminate Exonic Circular RNAs from Protein coding RNAs (mRNAs). Positional Statistics and entropy of dinucleotides within RNA sequence, and emission probability of RNA sequence were taken as discriminating features. Elman recurrent backpropagation neural network (RNN) was used as classifier for all individual features. Voted decision of multiple classifier system was used for final prediction of circular RNA and mRNA. Abstract: Circular RNAs are new class of stable non-coding RNAs, whose expressions are specific to tissues as well as developmental stages and reported to act as gene regulators. Conspicuous presences of some of them as biomarkers for cancers, aging etc. are well reported. Biogenesis of circular RNA competes with Pre-mRNA splicing using the same splicing machinery and gene loci. Also, some circular RNAs are reported to have open reading frames and internal ribosome entry site for ribosome binding, which increases the chance of overlapping features among circular and mRNA transcripts. Therefore, discriminating the Exonic circular RNAs and mRNAs solely through sequence properties is challenging. However, possible discriminating factors, such as, reports on non-canonical arrangement of exons in circular RNAs were cited. This study was dedicated to classify Circular RNAs from mRNAs by recruiting features extracted from sequences as well as predicted secondary structures and ANN classifier models for all these feature types. The features were statistics of di-nucleotide index, emission probability of RNA sequences and entropy of di-nucleotides. Finally a simple decision voting was applied to combine decisions obtained from multiple classifiers. After performing 10 fold cross validation we obtained average values of efficiency, sensitivity, specificity and Mathews correlation coefficient as 0.8374, 0.8544, 0.8203 and 0.6753 respectively. In the backdrop of few reports of identification of circular RNAs from constitutive exons and other long non-coding RNAs, this is the first report of discriminating exonic circular RNAs from mRNAs using sequence and sequence-derived properties.