Computational Prediction of Polycomb-Associated Long Non-Coding RNAs

• Published in: PloS one Vol. 7; no. 9; p. e44878
• Main Authors: Glazko, Galina V., Zybailov, Boris L., Rogozin, Igor B.
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 13.09.2012; Public Library of Science (PLoS)
• Subjects: Animals; Annotations; Base Sequence; Binding; Binding Sites; Bioinformatics; Biology; Chromatin - genetics; Chromatin - metabolism; Classifiers; Computational Biology; Computer applications; Conservation; Conserved sequence; DNA methylation; Epigenetics; Gene expression; Genomes; Humans; Hypotheses; Learning algorithms; Machine learning; Mice; Molecular Sequence Data; Polycomb group proteins; Polycomb Repressive Complex 2 - metabolism; Proteins; RNA, Antisense - genetics; RNA, Antisense - metabolism; RNA, Double-Stranded - genetics; RNA, Double-Stranded - metabolism; RNA, Long Noncoding - genetics; RNA, Long Noncoding - metabolism; Stem cells; United States > US; Arkansas
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0044878

Among thousands of long non-coding RNAs (lncRNAs) only a small subset is functionally characterized and the functional annotation of lncRNAs on the genomic scale remains inadequate. In this study we computationally characterized two functionally different parts of human lncRNAs transcriptome based on their ability to bind the polycomb repressive complex, PRC2. This classification is enabled by the fact that while all lncRNAs constitute a diverse set of sequences, the classes of PRC2-binding and PRC2 non-binding lncRNAs possess characteristic combinations of sequence-structure patterns and, therefore, can be separated within the feature space. Based on the specific combination of features, we built several machine-learning classifiers and identified the SVM-based classifier as the best performing. We further showed that the SVM-based classifier is able to generalize on the independent data sets. We observed that this classifier, trained on the human lncRNAs, can predict up to 59.4% of PRC2-binding lncRNAs in mice. This suggests that, despite the low degree of sequence conservation, many lncRNAs play functionally conserved biological roles.