NCLscan: accurate identification of non-co-linear transcripts (fusion,trans-splicing and circular RNA) with a good balance between sensitivity and precision

Abstract

Analysis of RNA-seq data often detects numerous ‘non-co-linear’ (NCL) transcripts, which comprised sequence segments that are topologically inconsistent with their corresponding DNA sequences in the reference genome. However, detection of NCL transcripts involves two major challenges: removal of false positives arising from alignment artifacts and discrimination between different types of NCL transcripts (trans-spliced, circular or fusion transcripts). Here, we developed a new NCL-transcript-detecting method (‘NCLscan’), which utilized a stepwise alignment strategy to almost completely eliminate false calls (>98% precision) without sacrificing true positives, enabling NCLscan outperform 18 other publicly-available tools (including fusion- and circular-RNA-detecting tools) in terms of sensitivity and precision, regardless of the generation strategy of simulated dataset, type of intragenic or intergenic NCL event, read depth of coverage, read length or expression level of NCL transcript. With the high accuracy, NCLscan was applied to distinguishing between trans-spliced, circular and fusion transcripts on the basis of poly(A)- and nonpoly(A)-selected RNA-seq data. We showed that circular RNAs were expressed more ubiquitously, more abundantly and less cell type-specifically than trans-spliced and fusion transcripts. Our study thus describes a robust pipeline for the discovery of NCL transcripts, and sheds light on the fundamental biology of these non-canonical RNA events in human transcriptome.

Reference-free andde novoIdentification of Circular RNAs

“…Current circRNA detection methods are incapable of distinguishing linear NCLs from circular NCLs (i.e., circRNAs) [15, 27] since they may share the same back-splicing junction [28]. In this study, the false positive rate caused by the linear intragenic chimeric transcripts was estimated as following: We identified 4,164 NCLs from the circRNA-enriched (poly(A) - /ribo - ) transcriptome using the tool NCLscan [15]. These NCLs were all circRNAs theoretically since the linear transcripts have been digested by RNase R during the library preparation.…”

“…In the same way, we identified 1,711 NCLs in the parallel total RNA transcriptome, which may consist of both linear NCLs and circRNAs. Of them, 686 were mutual to both transcriptomes, they were potential circRNAs [15]. Worthy of mention, three experiment-validated circRNAs (CDYL, GSE1, and ZNF609) were in the mutual list.…”

“…In recent years, a number of algorithms were introduced to detect circRNAs from RNA sequencing (RNA-seq) transcriptome in large scale. These algorithms, including find_circ [4], CIRCexplorer [11], circRNA_finder [12], CIRI [13], KNIFE [14], NCLscan [15], DCC [16], and UROBORUS [17], usually take the back-splice junction (BSJ) spanning reads as the critical clues in circRNA detection [4, 11, 13, 18]. They discover BSJs by either search of the wrong bridging events between the annotated exons (annotation dependent) or identification of potential splice patterns via mapping reads back to the reference genome or pseudo-sequences (annotation independent) [19].…”

Circular RNAs: Methodological challenges and perspectives in cardiovascular diseases

“…A stringent detection approach requires the use of a tool with high precision and therefore a controlled amount of false‐positives. NCLScan is a good example. This approach ensures a small list of strong candidates by stringent filtering and by imposing high detection thresholds.…”

PcircRNA_finder: a software for circRNA prediction in plants

“…Simulated RNA-Seq data (paired end reads, 100 ;bp and 6000 backsplicing reads for each sample) were generated by randomly choosing 200 chiastic transcripts based upon the Arabidopsis thaliana and rice genome annotations, respectively (Supplementary Data). The sensitivity, precision and sensitivity ;+ ;precision (a comprehensive value) (Chuang et al , 2016) was used to evaluate the performance of the three methods. The results indicate that PcircRNA_finder has a higher sensitivity (74–88%) than either find_circ or CIRCexplorer (each about 20%) and better precision (63–67%) compared to find_circ and CIRCexplorer, (72 and 100%, respectively) in the two test genomes (Supplementary Data).…”