In:
Bioinformatics, Oxford University Press (OUP), Vol. 36, No. 18 ( 2020-09-15), p. 4739-4748
Abstract:
CircRNAs are an abundant class of non-coding RNAs with widespread, cell-/tissue-specific patterns. Previous work suggested that epigenetic features might be related to circRNA expression. However, the contribution of epigenetic changes to circRNA expression has not been investigated systematically. Here, we built a machine learning framework named CIRCScan, to predict circRNA expression in various cell lines based on the sequence and epigenetic features. Results The predicted accuracy of the expression status models was high with area under the curve of receiver operating characteristic (ROC) values of 0.89–0.92 and the false-positive rates of 0.17–0.25. Predicted expressed circRNAs were further validated by RNA-seq data. The performance of expression-level prediction models was also good with normalized root-mean-square errors of 0.28–0.30 and Pearson’s correlation coefficient r over 0.4 in all cell lines, along with Spearman's correlation coefficient ρ of 0.33–0.46. Noteworthy, H3K79me2 was highly ranked in modeling both circRNA expression status and levels across different cells. Further analysis in additional nine cell lines demonstrated a significant enrichment of H3K79me2 in circRNA flanking intron regions, supporting the potential involvement of H3K79me2 in circRNA expression regulation. Availability and implementation The CIRCScan assembler is freely available online for academic use at https://github.com/johnlcd/CIRCScan. Supplementary information Supplementary data are available at Bioinformatics online.
Type of Medium:
Online Resource
ISSN:
1367-4803
,
1367-4811
DOI:
10.1093/bioinformatics/btaa567
Language:
English
Publisher:
Oxford University Press (OUP)
Publication Date:
2020
detail.hit.zdb_id:
1468345-3
SSG:
12
