Research Article
Assessing the impacts of various factors on circular
RNA reliability
Trees-Juen Chuang, Tai-Wei Chiang, Chia-Ying Chen
Circular RNAs (circRNAs) are non-polyadenylated RNAs with a
continuous loop structure characterized by a non-colinear
back-splice junction (BSJ). Although millions of circRNA can-
didates have been identified, it remains a major challenge for
determining circRNA reliability because of various types of
false positives. Here, we systematically assess the impacts of
numerous factors related to circRNA identification, conservation,
biogenesis, and function on circRNA reliability by comparisons of
circRNA expression from mock and the corresponding colinear/
polyadenylated RNA–depleted datasets based on three different
RNA treatment approaches. Eight important indicators of circRNA
reliability are determined. The relative contribution to variability
explained analyses reveal that the relative importance of these
factors in affecting circRNA reliability in descending order is the
conservation level of circRNA, full-length circular sequences,
supporting BSJ read count, both BSJ donor and acceptor splice
sites at the same colinear transcript isoforms, both BSJ donor and
acceptor splice sites at the annotated exon boundaries, BSJs
detected by multiple tools, supporting functional features, and
both BSJ donor and acceptor splice sites undergoing alternative
splicing. This study thus provides a useful guideline and an im-
portant resource for selecting high-confidence circRNAs for
further investigations.
DOI 10.26508/lsa.202201793 | Received 1 November 2022 | Revised 15
February 2023 | Accepted 15 February 2023 | Published online 27 February
2023
Introduction
CircRNAs are an emerging class of RNAs formed by cis-backsplicing
with a covalently closed loop structure that are endogenously
expressed as non-polyadenylated circular molecules (1, 2). Such a
loop structure is characterized by a non-colinear (NCL) back-splice
junction (BSJ) between a downstream splice donor site and up-
stream splice acceptor site. Despite most circRNAs are expressed at
a much lower level compared with linear mRNAs (3, 4, 5), some are
highly expressed (3, 6), even more abundant than their colinear
counterparts (7, 8). In addition, circRNAs are more stable than their
corresponding colinear mRNA isoforms (1, 9). Some circRNAs were
Genomics Research Center, Academia Sinica, Taipei, Taiwan
Correspondence: trees@gate.sinica.edu.tw.
© 2023 Chuang et al.
further demonstrated to be evolutionarily conserved across spe-
cies (3, 10, 11, 12). Accumulating evidence demonstrated that
circRNAs can participate in various aspects of biological processes,
including competing with their host gene expression and splicing,
regulating activities of miRNAs and RNA-binding proteins
(RBPs), and acting as translation templates (2, 13). CircRNAs were
reported to be especially abundant in brain and neuronal tissues
(11, 14) and regulatory important in nervous system development
and aging (15). Loss or dysregulation of circRNAs may affect brain
function (12, 15, 16, 17), indicating the roles of circRNAs in patho-
genesis and progression of neurological diseases. In cancer studies,
alternation of circRNA expression was shown to affect cell apoptosis,
invasion, migration, and proliferation, revealing the potential of
circRNAs for serving as biomarkers and therapeutic targets in cancer
(18, 19). These findings suggest that circRNAs are an ancient and fine-
tuned class of functional transcripts.
Varying bioinformatic tools have been developed to identify BSJs
using high-throughput transcriptome sequencing (RNA-seq) data
(20), leading to a tremendous number of potential circRNA candidates in
human and other species (21, 22, 23). However, circRNA detections are
hampered by various types of false positives. An NCL junction that
originates from sequencing error, in vitro artifact, ambiguous alignment,
trans-splicing, or genetic rearrangement is often misjudged as a BSJ
event by computational detections (1, 2). This reflects the fact of limited
overlap among the circRNA candidates identified by different tools (1, 24,
25, 26, 27) or databases (23), highlighting the issue of circRNA reliability.
Although millions of circRNA candidates were deposited in varied da-
tabases, only a very limited number of candidates have been carefully
curated to date (23). It requires a guideline for evaluating the reliability of
the detected circRNA candidates before performing time- and cost-
consuming experimental validations.
Because circRNAs are NCL RNAs and lack 39polyadenylated tails,
many circRNA candidates were detected in RNA-seq datasets de-
rived from total RNAs (the ribosomal RNA–depleted RNAs without
poly(A)-selection; “mock RNAs” for simplicity), which contain RNA-
seq reads from both colinear and NCL RNAs (28). Because most
colinear RNAs are digested by RNase R (3, 28, 29, 30), a detected
circRNA candidate is more likely to be real if its normalized BSJ read
count is enriched or not reduced after the treatment. Therefore,
comparisons of BSJ read counts detected from mock and RNase
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 1 of 16
R–treated RNA datasets were often employed to evaluate the re-
liability of the detected circRNAs (24, 25, 26, 31). However, in some
cases, such a mock-treated comparison approach may not be ef-
ficient for circRNA detection because of two inherent limitations.
First, some colinear RNAs containing highly structured 39 ends or
G-quadruplexes are also resistant to RNase R (32). Second, some
circRNAs may be susceptible to RNase R and digested by this exonuclease
(33). Accordingly, in addition to RNase R–treated RNAs, we considered
two other RNA treatments, RNAs treated with coupling A-tailing
and RNase R digestion and RNAs treated with depletion of both
ribosomal RNAs and polyadenylated RNAs, for mock-treated
comparisons. The former could efficiently digest most of the
RNase R–resistant colinear RNAs stated above (32), and the latter
could deplete most polyadenylated RNAs and retain circRNAs
susceptible to RNase R (33). On the basis of comparisons of
normalized BSJ read counts detected from mock and the cor-
responding colinear/poly(A)-depleted RNA datasets from three
different RNA treatments, we thus presented a systematical
workflow to evaluate the impacts of various factors related to circRNA
identification, conservation, biogenesis, and function on circRNA
reliability. After determining statistically important indicators of
circRNA reliability, we further assessed the relative influence of
these important indicators on circRNA reliability. Collectively,
we provided a useful guideline for selecting high-confidence
circRNAs. All of the information related to circRNA identification,
conservation, biogenesis, and function was freely available.
Results
Potential false positives in the circRNA database
As shown in our analysis flow (Fig 1A), we first extracted circRNA (or
BSJ) candidates from the circAtlas database 2.0 (22) (580,654 can-
didates; see also Table S1). Like previous reports (1, 24, 25, 26, 27, 34),
a considerable percentage (36.5%) of the examined circRNA can-
didates were detected by only one tool (Fig 1B). Such great dis-
crepancies in the identification results among tools reflected the
uncertainty of the computationally detected circRNA events (1, 24, 25,
26, 27). One cause of the false-positive calls may be due to alignment
ambiguity with an alternative colinear explanation or multiple hits
when detecting BSJs (1, 26, 35). Of the 580,654 circAtlas circRNAs, we
found that 17.3% (100,183 circRNAs; Fig 1C) were potential false positives
arising from an alternative colinear explanation (Fig S1) or multiple hits
(see the Materials and Methods section). It was reported that the
circRNAs detected by multiple tools tended to be more reliable than
tool-specific circRNAs (36, 37). As expected, the percentages of circRNAs
derived from ambiguous alignments significantly decreased with in-
creasing the numbers of the detected tools (Fig 1D), reflecting
the positive correlation between the level of circRNA reliability and
the number of circRNA detection tools. For accuracy, we excluded the
circRNA candidates arising from potential alignment ambiguity
(100,183 events) and only considered the remaining circRNAs (480,471
events) for the following analyses of circRNA reliability.
Most circRNA candidates were detected from RNA-seq datasets
derived from mock samples (the ribosomal RNA–depleted RNAs
Important indicators of circRNA reliability Chuang et al.
without poly(A)-selection), which contain RNA-seq reads from both
circRNAs and colinear/polyadenylated RNAs (28). We employed
comparisons of normalized BSJ read counts detected from mock
and the corresponding colinear/poly(A)-depleted datasets to de-
termine high-confidence circRNAs. For comprehensively assessing
circRNA reliability and increasing the robustness of our analyses,
we extracted 19 mock-treated sample pairs of different samples or
studies from public datasets, which included three types of RNA
treatment approaches: RNase R approach (group 1), A-tailing RNase
R approach (group 2), and non-poly(A) approach (group 3) (see
Tables 1 and S2). We examined the 480,471 circRNAs in all the mock
samples (see the Materials and Methods section) and found that a
considerable percentage of circRNAs (38–80%) were supported by only
one BSJ read (Fig 1E). To minimize potential false positives, only the
circRNAs expressed in the mock samples with the supporting BSJ read
count ≥ 2 were considered in the analysis for each mock-treated sample
pair. It was presumed that the circRNAs not depleted after the treatment
(“not-depleted circRNAs”) were more likely to be bona fide circRNAs
than those depleted because of the treatment (“depleted circRNAs”). We
then compared not-depleted circRNAs with depleted ones in terms of
various factors related to circRNA identification, conservation, biogen-
esis, and function (Table S3), and thereby accessed the importance of
these factors in affecting circRNA reliability.
Factors related to circRNA identification
Regarding the circRNAs identified in the mock samples, the impacts
of the factors related to circRNA identification (i.e., BSJ read count,
number of detected tools, and evidence of full-length circular
sequence) on circRNA reliability were assessed. First, we found
that the supporting BSJ read counts were significantly higher in
not-depleted circRNA candidates than in depleted ones in all
mock-treated sample pairs examined (all FDR < 0.05; Fig 2A). The
percentages of non-depleted circRNAs increased with increasing the
supporting BSJ read counts (Fig 2A). Second, we examined the in-
fluences of BSJs detected by multiple or single circRNA detection
tools on circRNA reliability and found that the former were significantly
enriched for not-depleted circRNAs (Fig 2B). Generally, the percentages of
non-depleted circRNAs increased with increasing numbers of the de-
tected tools (Fig 2B). Third, we examined the influences of BSJs with or
without the evidence of full-length circular sequences reconstructed by
CIRI-full (43) on circRNA reliability and showed that the former were
significantly enriched for not-depleted circRNAs, regardless of the RNA
treatment approaches (Fig 2C). Because the performance of CIRI-full, a
short read–based approach, for identifying full-length sequences of
circRNAs is hampered by the length of short reads (43), we also extracted
the full-length circRNA candidates identified by circFL-seq based on
nanopore long reads (44). Such a long read–based approach provided the
direct support of full-length circular sequences from single-molecule long
reads. Similarly, the BSJ events with the evidence of full-length circular
sequences showed a significant enrichment for not-depleted circRNAs
in all mock-treated sample pairs examined (Fig 2C).
Factors related to circRNA conservation
We then examined the relationship between circRNA reliability and
conservation patterns at the three levels: species, tissues, and
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 2 of 16
Figure 1. The assessment of the impacts of various features on circRNA reliability.
(A) Flowchart of the overall analyses. (B) Distribution of the extracted circAtlas circRNA (BSJ) candidates (580,654 candidates) identified by 1, 2, 3, or 4 circRNA detection
tools. (C) 580,654 candidates derived from potential alignment ambiguity (with an alternative colinear explanation or multiple hits). (D) Comparisons of the percentages
of circRNA candidates derived from potential alignment ambiguity for the candidates detected by 1, 2, 3, or 4 tools. (E) Comparisons of normalized numbers of circRNA
candidates with supporting BSJ read count = 1 or ≥ 2 in all extracted mock samples of the 19 mock-treated sample pairs. For each sample, the percentage of circRNA
candidates supported by one BSJ read is shown.

individuals (or samples). We found that the conservation levels of
circRNAs at the three levels (Fig 3A–D) were all significantly higher in
not-depleted circRNAs than in depleted ones, regardless of the RNA
treatment approaches. The percentages of non-depleted circRNAs
increased with increasing numbers of the conserved species (Fig
3A), tissues (Fig 3B), and samples (Fig 3D). Of note, in contrast with
the number of tissues (Fig 3B), the levels of tissue specificity index
were significantly lower in not-depleted circRNA than in depleted
ones (Fig 3C). We further showed that the evolutionary rates (as
determined by phyloP (45) or phastCons (46)) of the sequences
around the BSJs (see see the Materials and Methods section) were
generally higher in not-depleted circRNAs than in depleted ones
(Fig S2). These results suggested that the BSJ events widely detected
across species/samples were more reliable than those detected in
a limited number of species/samples.
Factors related to circRNA biogenesis
We proceeded to examine the relationships between circRNA re-
liability and the factors related to circRNA biogenesis. Because

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 3 of 16

Table 1. Summary of RNA-seq data used in this studya.
Approach (mock versus treated RNAs)b
Group 1: RNase R approach
rRNA- versus RNase R+
rRNA- (KCl) versus RNase R+ (KCl)
rRNA- versus RNase R+ (KCl)
rRNA- versus RNase R+ (LiCl)
Group 2: A-tailing RNase R approach rRNA- (LiCl) versus A-
tailing/RNase R+ (LiCl)
rRNA- versus A-tailing/RNase R+ (LiCl)
Group 3: non-poly(A) approach rRNA- versus p(A)-
a
The detailed information is given in Table S2.
b
rRNA-, RNase R+, A-tailing/RNase R+ (LiCl), and p(A)-represented rRNA-depleted RNAs, rRNA-depleted RNAs with RNase R digestion, rRNA-depleted RNAs
treated with coupling A-tailing and RNase R digestion in LiCl-containing buffer, and rRNA-/poly(A)-depleted RNAs.
circRNAs were suggested to be generated by canonical spliceo-
somal machinery (1, 5, 47, 48, 49), we examined whether the BSJs
of not-depleted circRNA candidates tended to agree to well-
annotated exon boundaries of colinear transcripts or be subject
to alternative splicing (AS) based on previously annotated colinear
transcripts (e.g., the Ensembl annotation) as compared with those
of depleted circRNA ones. Indeed, the BSJs that agreed to annotated
exon boundaries (either donor or acceptor splice sites) showed a
significant enrichment for not-depleted circRNAs, regardless of the
RNA treatment approaches (Figs 4A and S3A). The BSJs with both
donor and acceptor splice sites matching annotated exon
boundaries from the same colinear transcript isoforms were also
enriched for not-depleted circRNAs (Fig 4B). We also observed that
the splice site strength of BSJs was generally stronger in not-
depleted circRNAs than in depleted ones, regardless of the tools
used for estimating the splice site strength (Fig S3B). Regarding the
BSJs matching previously annotated exon boundaries of colinear
transcripts, the BSJs (donor or acceptor splice sites) that were
subject to AS were significantly enriched for not-depleted circRNAs
across the mock-treated sample pairs examined (Fig S3C). Such a
trend exhibited particularly enriched when both donor and ac-
ceptor splice sites of the BSJs were subject to AS (Fig 4C). These
observations support that most circularized exons are recognized
by the spliceosome during canonical splicing.
Next, previous studies demonstrated that backsplicing can be
facilitated by RCSs residing in the sequences flanking circularized
exons (3, 8, 17, 29) and affected by the competition of RCSs across
flanking regions (RCSacross) or within individual regions (RCSwithin) (29). We
thus asked whether the BSJs with RCSacross in flanking regions were
enriched for not-depleted circRNA candidates. However, we did not
observe this trend (Fig 4D). We further examined the numbers of
(RCSacross - RCSwithin) for the detected circRNAs and found no signif-
icant differences between not-depleted and depleted circRNAs in
terms of the percentages of the BSJs with (RCSacross - RCSwithin) ≥ 1
Important indicators of circRNA reliability Chuang et al.
Sample pair ID Resource (ref.)
HeLa_1 PRJNA266072 (38)
Hs68 SRP011042 (3)
HeLa_2, HEK293 PRJNA266072 (39)
HeLa_3, K562 PRJNA231721 (40)
Total_brain, cerebellum, and blood PRJNA646808 (34)
HeLa_KCl PRJNA541935 (32)
HEK293T_KCl and SH-SY5Y_KCl PRJNA752779 (41)
HEK293T_LiCl and SH-SY5Y_LiCl PRJNA752779 (41)
HeLa_A PRJNA541935 (32)
HEK293T_A and SH-SY5Y_A PRJNA752779 (41)
GM12878_pA- and K562_pA- ENCODE (42)
(Fig 4E). In addition, a number of RBPs were demonstrated to serve as
a role for regulating circRNA biogenesis by binding to cis elements in
the sequences flanking circularized exons (1, 2). We then examined
whether the factor of BSJs with RBPs binding to the flanking regions
showed enrichment for not-depleted circRNAs and found that this
factor was slightly enriched for not-depleted circRNAs only (Fig 4F). We
also examined the BSJs with RCSacross or RBP-binding sites in flanking
regions and did not observe a significant enrichment for not-depleted
circRNAs (Fig 4G). Moreover, considering the minimum distance of the
detected RBP-binding sites to the BSJs, the distance was only slightly
shorter in not-depleted circRNAs than in depleted ones (Fig 4H). These
results suggested that the factors of RCS or RBPs binding to the flanking
regions were not good indicators for evaluating circRNA reliability.
Factors related to circRNA function
We then examined the impact of functional features (miRNA binding, RBP
binding, and seven types of evidence for circRNA coding potential) on
circRNA reliability (see the Materials and Methods section). For each BSJ
event, because the circularized sequence is shared with its colinear
counterpart except for the BSJ donor and acceptor splice sites, we con-
sidered the predicted miRNA/RBP-binding sites and ORFs spanning the
BSJ. Integration of these nine types of functional features, we observed that
numbers of supporting functional features were significantly higher in not-
depleted circRNAs than in depleted ones in all mock-treated sample pairs
(all FDR < 0.05; Fig 5). The percentages of non-depleted circRNAs increased
with increasing numbers of supporting functional features (Fig 5). These
results suggested that the strength of functional evidence for circRNA
candidates were positively correlated with the level of circRNA reliability.
Relative effect of each individual factor on circRNA reliability
According to the above assessment results, we observed eight
important indicators of circRNA reliability, each of which can
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 4 of 16
Figure 2. Impact of factors related to circRNA identification on circRNA reliability.
(A) Supporting BSJ read count, (B) number of detected tools, and (C) evidence of full-length circular sequence. For (B, C), the odds ratios, which were determined using
two-tailed Fisher’s exact test, represented the ratios of the occurrences of (B) circRNAs detected by multiple tools or (C) circRNAs with the evidence of full-length circular
sequence for non-depleted circRNAs to the occurrences of those for depleted circRNAs. The dashed lines represented odds ratio = 1. (A, B) For the bottom panels of (A, B),
the correlations between percentages of non-depleted circRNAs and (A) supporting BSJ read count or (B) number of detected tools are shown. For (C), the evidence of
full-length circular sequence was supported by CIRI-full (a short read-based approach, top) or circFL-seq (a long read-based approach, bottom). (A, B, C) P-values were
determined using Wilcoxon rank-sum test (WRST; greater or less; (A)) or two-tailed FET (B, C) and FDR adjusted across 19 mock-treated sample pairs for each examined
factor using Benjamini–Hochberg correction. For (A), the Wilcoxon effect sizes and the corresponding 95% confidence intervals are plotted (see also Table S4). The
number of mock-treated sample pairs that passed the statistical significance tests with FDR < 0.05 (or −log10(FDR) > 1.3) are represented in curly brackets.

effectively distinguish between non-depleted and depleted
circRNAs in all mock-treated sample pairs examined. The eight
factors were (i) supporting BSJ read count; (ii) BSJs detected by
multiple tools; (iii) full-length circular sequences; (iv) conservation
level of circRNA (number of samples observed the circRNAs); (v)
both BSJ donor and acceptor splice sites at the annotated exon
boundaries; (vi) both BSJ donor and acceptor splice sites at the
same colinear transcript isoforms; (vii) both BSJ donor and
acceptor splice sites undergoing annotated AS events; and (viii)
supporting functional features. We then employed a general-
ized linear model (GLM) with the eight factors and measured
the relative contributions to variability explained (RCVE) (50) to
evaluate the relative effect of individual factor in affecting
circRNA reliability for each mock-treated sample pair (see the
Materials and Methods section) (Fig 6A and Table S4). We
ranked the RCVE scores for each mock-treated sample pair and
found that on average the most important features in descending
order were the conservation level of circRNA, full-length circular
sequences, supporting BSJ read count, both BSJ donor and acceptor
splice sites at the same colinear transcript isoforms, supporting
functional features, both BSJ donor and acceptor splice sites at the
annotated exon boundaries, BSJs detected by multiple tools, and
both BSJ donor and acceptor splice sites undergoing annotated AS
events (Fig 6B).
Particularly, our result revealed that the conservation level of
circRNA (number of samples observed the circRNAs) is the most

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 5 of 16

Figure 3. Impact of factors related to circRNA conservation on circRNA reliability.
(A, B, C, D) Impact of conservation factors at the (A) species, (B, C) tissue, and (D) individual (or sample) levels on circRNA reliability. For the bottom panels of (A, B, and D),
the correlations between percentages of non-depleted circRNAs and (A) number of conserved species, (B) number of conserved tissues, or (C) number of conserved
samples are shown. P-values are determined using WRST (greater or less) and FDR adjusted across 19 mock-treated sample pairs for each examined factor using
Benjamini–Hochberg correction. The Wilcoxon effect sizes and the corresponding 95% confidence intervals are plotted (see also Table S4). The number of mock-treated
sample pairs that passed the statistical significance tests with FDR < 0.05 (or −log10(FDR) > 1.3) are represented in curly brackets.

dominant determinant of circRNA reliability common to all the
examined mock-treated sample pairs except for the sample pair in
blood (Fig 6B). Of note, the samples considered were the integration
of samples across species, tissue, and individuals in circAtlas. It is
known that circRNAs are often expressed in a cell type-/tissue-
specific manner (4, 51). Actually, we found that most (61%; 293,152
circRNAs) of the 480,471 circAtlas circRNAs were detected in one
sample only (Table S3). We were curious about whether the positive
correlation between the percentage of not-depleted circRNAs and the
number of samples observed the circRNAs held when considering
one cell type/tissue only. Of the examined mock-treated sample
pairs, the RNA-seq data of five pairs (HeLa_1, HeLa_2, HeLa_3,
HeLa_KCl, and HeLa_A) were all derived from HeLa cells, which were
generated from different studies or conditions (Table S2). The five
HeLa-based RNA-seq datasets can be regarded as replicates. As
expected, the percentage of not-depleted circRNAs was strongly
positively correlated with the number of samples observed the
circRNAs in all the five HeLa-based pairs (Fig 6C, left). The similar
trend was observed for the two K562-based pairs (Fig 6C, middle).
We also examined the HeLa_1 and HeLa_2 mock-treated sample

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 6 of 16

Figure 4. Impact of factors related to circRNA biogenesis on circRNA reliability.
(A) Both BSJ sites at annotated boundaries, (B) both BSJ sites at the same isoform, (C) both BSJ sites undergoing alternative splicing (AS), (D) BSJs with #RCSacross > 0,
(E) BSJs with #(RCSacross - RCSwithin) > 0, (F) BSJs with RBPs binding to the flanking regions, (G) BSJs with #RCSacross > 0 or RBPs binding to the flanking regions, and
(H) minimum distance of RBP-binding sites to BSJs. For (A, B, C, D, E, F, G), the odds ratios represented the ratios of the occurrences of the examined factors of (A, B, C, D, E, F,
G) for non-depleted circRNAs to the occurrences of those for depleted circRNAs. The dashed lines represented odds ratio = 1. Odds ratios and P-values were determined
using two-tailed FET. P-values were FDR adjusted across 19 mock-treated sample pairs for each examined factor using Benjamini–Hochberg correction. For (H), P-values
were determined using WRST (greater or less) and FDR adjusted across 19 mock-treated sample pairs for each examined factor using Benjamini–Hochberg correction. The
Wilcoxon effect sizes and the corresponding 95% confidence intervals are plotted (see also Table S4). The number of mock-treated sample pairs that passed the
statistical significance tests with FDR < 0.05 (or −log10(FDR) > 1.3) are represented in curly brackets.

pairs, which were two replicates generated from the same
group but different studies (Table S2) and showed the similar
results (Fig 6C, right). Furthermore, regarding the mock samples
of the 19 mock-treated sample pairs, we extracted the HeLa-
specific circRNAs that were detected in the examined HeLa
samples only but not in the non-HeLa cell lines/tissues. For
each of the five HeLa-based mock-treated sample pairs, the HeLa-
specific circRNAs were divided into circRNAs detected in single replicate
only and circRNAs detected in multiple replicates. Our results revealed
that the percentages of not-depleted circRNAs were significantly
higher in circRNAs detected in multiple HeLa replicates than in
circRNAs detected in single replicate only for all five HeLa-based

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 7 of 16

Figure 5. Impact of functional features on circRNA reliability.
Nine types of functional features (see the Materials and Methods section) for circRNAs are examined. For the bottom panel, the correlations between percentages of
non-depleted circRNAs and number of supporting functional features are shown. P-values are determined using WRST (greater or less) and FDR adjusted across 19 mock-
treated sample pairs for each examined feature using Benjamini–Hochberg correction. The Wilcoxon effect sizes and the corresponding 95% confidence intervals are
plotted (see also Table S4). The number of mock-treated sample pairs that passed the statistical significance tests with FDR < 0.05 (or −log10(FDR) > 1.3) are represented
in curly brackets.

pairs (Fig 6D, left). Such a trend was held for the K562-based pairs
(Fig 6D, right).
Discussion
The reliability of computationally detected circRNAs is not guar-
anteed because of varied types of false positives. It is particularly
difficult to pick out spurious BSJ (or circRNA) events derived from
in vitro artifacts among a tremendous number of previously
identified NCL junctions. Such in vitro artifacts may frequently arise
from template switching by random hexamer primed reverse
transcriptase (RT) reactions (52), which can switch from one tem-
plate to another with short homologous sequences at the NCL
junction sites (53, 54). Of note, some NCL junctions passing RT-
based validations were finally confirmed to be generated from
in vitro artifacts by further validations (10). As RNA-seq data are
generated from RT-based sequencing strategies, it is unavoidable
that a number of spurious NCL events masquerade as genuine BSJs
because of such RT-based artifacts (1). We asked whether the
approaches based on comparisons of paired mock and treated
samples can reflect circRNA reliability to a certain extent. We
emphasized that, for the robustness of our analyses, 19 mock-
treated sample pairs of different samples/studies based on three
types of RNA treatment approaches (Table 1): RNase R approach
(group 1), A-tailing RNase R approach (group 2), and non-poly(A)
approach (group 3) were performed for the mock-treated com-
parisons. Because highly structured 3’ ends or G-quadruplexes may
affect the effect of the RNase R treatment (32), we also examined
whether the presence of G-quadruplexes across the BSJs affected
the percentages of not-depleted circRNAs and found no significant
differences between circRNAs with and without removing
G-quadruplexes across BSJs (Fig S4 and Table S3). Moreover, we
extracted BSJs from two high-confidence circRNA datasets, in which
the BSJ events had passed multiple experimental validations. For
the first dataset (8), BSJs were supported by both avian myelo-
blastosis virus- and Moloney murine leukemia virus-derived
RNA-seq reads (designated as “RT-independent circRNAs”). RT-
independent circRNAs are regarded to be highly accurate be-
cause comparisons of different RTase products can effectively
detect RT-based artifacts (1, 10, 52, 53). For the second dataset (55),
BSJs (or their functions) were previously validated by RT-based
experiments and at least one type of non-RT–based experiments
simultaneously in at least one human tissues or cell lines (des-
ignated as “RT-/non-RT–validated circRNAs”; see the Materials and
Methods section). We observed that the percentages of both RT-
independent (Fig 7A) and RT-/non-RT–validated (Fig 7B) circRNAs
were significantly higher in not-depleted circRNAs than in depleted
circRNAs in all mock-treated sample pairs examined. Moreover, we
compared circAtlas circRNAs with circRNAs from eight other pub-
licly accessible circRNA databases and identified circAtlas-specific
circRNAs. We presumed that circAtlas-specific circRNAs were less
probably to be reliable as compared with circRNAs collected in
multiple databases. Indeed, the percentages of circAtlas-specific
circRNAs were significantly lower in not-depleted circRNAs than in
depleted circRNAs (Fig 7C).
For the relative influence of each individual factor on circRNA
reliability, our RCVE analysis revealed that “conservation level of
circRNA” and “full-length circular sequences” were the top two

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 8 of 16

Figure 6. Assessment of relative influence of each individual factor on circRNA reliability.
(A) RCVE scores of the examined factors for each mock-treated sample pair. The length of each color segment in each bar represents the RCVE score of the corresponding
examined factor. (B) Ranking (top) and average ranking (bottom; see also the numbers in the parentheses) of the RCVE scores of the examined factors. (C) Correlations between the
percentages of not-depleted circRNAs and number of samples observed the circRNAs in HeLa-based (left and right) or K562-based (middle) mock-treated sample pairs. Of note, the
HeLa_1 and HeLa_2 mock-treated sample pairs were generated from the same group but different studies. (D) Comparisons of percentages of not-depleted circRNAs for HeLa-
specific circRNA (left) or K562-specific circRNAs (right) detected in single replicate only or multiple replicates. All P-values are determined using two-tailed FET.

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 9 of 16


Figure 7. Robustness analyses of the approaches based on comparisons of
paired mock and treated samples for assessing circRNA reliability.
(A, B, C) Comparisons of the percentages of (A) RT-independent circRNAs, (B) RT-/
non-RT–validated circRNAs, and (C) circAtlas-specific circRNAs in the not-depleted
and depleted circRNAs for all mock-treated sample pairs examined. RT-
independent and RT-/non-RT–validated circRNAs represented high-confidence
circRNAs (see text and the Materials and Methods section). CircAtlas-specific
circRNAs were the circAtlas circRNAs that are not observed in eight other publicly
accessible circRNA databases (see the Materials and Methods section). P-values
were determined using two-tailed FET and FDR adjusted across 19 mock-treated
sample pairs for each examined feature using Benjamini–Hochberg correction.
*FDR < 0.05. **FDR < 0.01. ***FDR < 0.001.
dominant determinants of circRNA reliability (Fig 6B). For the most
important factor in affecting circRNA reliability, because template
switching events occur randomly during reverse transcription
in vitro, a circRNA candidate observed in multiple samples or
replicates is less likely to be generated from such an in vitro artifact.
For the second most important factor in affecting circRNA reliability,
full-length circular sequences can be reconstructed by Illumina
short RNA-seq reads with the reverse overlap (RO) feature (43) or
identified by single-molecule long reads (44). With the support of
Important indicators of circRNA reliability Chuang et al.
RO-merged reads, the BSJ events are less probable to originate
from template switching because both ends of the paired-end
reads are sequencing from the same fragment and reversely
overlapped with each other by alignment. Similarly, with the direct
support of full-length circular sequences from single-molecule
long reads, the nanopore-based full-length circRNAs are less
likely to be artifacts derived from template switching.
Our analyses also suggested that “supporting BSJ read count,”
“supporting functional features,” and “BSJs detected by multiple
tools” were good indicators for assessing circRNA reliability. Be-
cause the circularized sequence is shared with its colinear coun-
terpart except for the BSJ sites, the supporting BSJ read is a unique
evidence of charactering a given circRNA event. It is known that
most circRNAs are expressed at a low level (3, 4, 5); particularly, a
considerable number of detected circRNA candidates were sup-
ported by one BSJ read only (e.g., Fig 1E). It is challenging to de-
termine the reliability of such low-abundance circRNAs. Detection
tools often set a threshold of supporting BSJ read count to minimize
false positives. For example, circRNA_finder (56) and segemehl (57)
filter circRNA candidates with a BSJ read count of at least five. In
addition to BSJ reads, the functional features of the predicted
miRNA/RBP-binding sites or ORFs spanning the BSJs implies the
relevance of the BSJ events to functional importance. These fea-
tures provide another unique support of circRNA reliability. Re-
garding the factor of “BSJs detected by multiple tools,” a previous
study reported that combining the results of multiple tools may
compensate for the possible weak points of each single tool, and
thereby increase the confidence of the detected circRNAs (58). This
also reflects our observation of the negative correlation between
the percentages of false-positive BSJs derived from alignment
ambiguity and the numbers of the detected tools (Fig 1D).
In terms of circRNA biogenesis, we showed that the factors of
“both BSJ donor and acceptor splice sites at the same colinear
transcript isoforms,” “both BSJ donor and acceptor splice sites at
the annotated exon boundaries,” and “both BSJ donor and acceptor
splice sites undergoing annotated AS events” were important for
determining circRNA reliability. A possible reason is that circRNAs
are also generated by canonical spliceosomal machinery (1, 47, 48,
49). Previous studies also reported that NCL transcript products
(including circRNAs, fusion events, and trans-spliced transcripts)
not matching well-annotated exon boundaries were more likely to
be in vitro artifacts (53, 59, 60). Particularly regarding the transcripts
generated during the transcriptional process, in addition to cis-
backsplicing, an observed intragenic NCL junction may also arise
from trans-splicing (1, 10). Although most of the detected NCL
events may be derived from cis-backsplicing, previous studies
based on comparisons of poly(A)-selected and poly(A)-depleted
RNA-seq data suggested that a considerable percentage of de-
tected NCL junctions were derived from trans-spliced RNAs (8, 61).
Actually, a number of NCL junctions detected in human cells were
experimentally validated to be generated by trans-splicing rather
than cis-backsplicing (10, 53, 62, 63). As trans-splicing occurs be-
tween two or more separate precursor mRNAs (64, 65), it is possible
that an observed NCL junction with its two splice sites from dif-
ferent colinear transcript isoforms is derived from a trans-splicing
event. We observed that the full-length circRNAs with the support of
RO-merged short reads or single-molecule long reads, which were
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 10 of 16
principally not generated from trans-splicing, were significantly the best of our understanding, this study, for the first time, presents
enriched for the BSJs with donor and acceptor splice sites at the a useful guideline to systematically assess circRNA reliability with
same colinear transcript isoforms (Fig S5). simultaneously accounting for varied types of factors, facilitating
On the other hand, we showed that the factors related to RCS and further functional investigation of this important class of non-
RBPs binding to the flanking regions of BSJs were not good indi- canonical transcripts.
cators for assessing circRNA reliability. Our observations reflect a
previous notion that the existence of RCS is neither sufficient nor
necessary for cis-backsplicing (2). Although some circRNA cases
were demonstrated to be facilitated by base-paring between RCSs
Materials and Methods
at two flanking intronic sequences of the BSJs (3, 8, 29), only a
CircRNA candidates and RNA-seq data collection
limited number of RCS-circRNA correlations were experimentally
confirmed. Multiple RNA parings derived from different RCS pairs
As shown in Fig 1A, human circRNA candidates examined (Table S1)
across flanking introns or within individual flanking introns can
were extracted from circAtlas 2.0 (22), which contained 580,654
compete against each other and lead to the reduction of circRNA
human circRNA candidates based on 240 samples collected from 19
production (29, 47, 66) or alternative cis-backsplicing (30, 67), further
tissues across seven vertebrate species. The related information
complicating the determination and validation of the RCS-circRNA
such as circRNA identification factors (e.g., circRNA detection tools
correlations. In addition, although repetitive elements abundantly
and full-length circular sequences) and conservation factors (e.g.,
emerge in mammalian genomes, which contribute to RNA pairing,
conservation of circRNAs across species, tissues, and samples)
circRNAs observed in non-mammalian species (e.g., Drosophila
were also downloaded from circAtlas 2.0. CircAtlas circRNAs were
melanogaster (56) and Oryza sativa (68)) are often lacking the feature
identified by CIRI2 (38, 72), find_circ (28), CIRCexplorer (73), or DCC
of RCS. For the associations between RBPs and circRNA formation,
(74). The full-length circular sequences were reconstructed by CIRI-
different RBPs may play opposite roles (up- or down-regulation) in
full (43). All the analyses were based on the human reference
regulating circRNA expression. Some RBPs, such as FUS (69) and
genome (GRCh38) and the Ensembl annotation (version 100). The
ADARs (70), were even showed to mediate circRNA production in a
circRNA candidates that were potentially derived from alignment
bidirectional manner. Because different RBPs may have overlapping
ambiguity (circRNA candidates had an alternative colinear expla-
capacities for binding to RCSs, it requires further investigations to
nation or multiple matches against the human genome) were
understand the joint (coordinate or competitive) effects of different
detected by NCLcomparator (35) with default parameters and were
RBPs on the regulations for each individual circRNA (2, 71).
not included in the subsequent analyses for accuracy (Fig 1A). After
This study systematically assessed the impacts of a dozen
that, 480,471 circAtlas circRNAs were retained. The remaining RNA-
factors related to identification, conservation, biogenesis, and
seq reads were aligned against the human reference genome using
function on circRNA reliability on the basis of comparisons of mock-
STAR with default parameters (75). For each circRNA (BSJ), we
treated sample pairs from three different RNA treatment ap-
generated a pseudo sequence by concatenating the sequences
proaches. Of the examined factors, eight were shown to be the
flanking the BSJ (within −50 nucleotides of the donor site to +50
important indicators of circRNA reliability in all mock-treated
nucleotides of the acceptor site) and aligned all STAR-unmapped
sample pairs examined, regardless of the RNA treatment ap-
reads against the 100-bp BSJ-based pseudo sequence using BWA
proaches. We assessed the relative influence of each individual
with default parameters (76). A STAR-unmapped read was deter-
factor on circRNA reliability and suggested that the most important
mined as a BSJ read if the read matched to ≥80% of the BSJ-based
factors in descending order were the conservation level of circRNA,
pseudo sequence and spanned the junction boundary by ≥10 bp on
full-length circular sequences, supporting BSJ read count, both BSJ
both sides of the BSJ.
donor and acceptor splice sites at the same colinear transcript
The mock-treated sample pairs (19 pairs) were extracted from
isoforms, supporting functional features, both BSJ donor and ac-
publicly accessible databases and categorized into three groups
ceptor splice sites at the annotated exon boundaries, BSJs detected
according to the RNA treatment approaches (see Tables 1 and S2):
by multiple tools, and both BSJ donor and acceptor splice sites
undergoing annotated AS events. All the circAtlas circRNA candi- Group 1: RNase R approach, mock sample versus RNase R–treated
dates and the corresponding factors examined were provided sample (14 pairs).
(Table S3). Importantly, we found a remarkably positive correlation Group 2: A-tailing RNase R approach, mock sample versus A-tailing
between number of supporting factors and the percentage of high- RNase R–treated (RNAs treated with coupling A-tailing and RNase R
confidence circRNAs (i.e., RT-independent and RT-/non-RT– digestion) sample (three pairs).
validated circRNAs) and a remarkably negative correlation between Group 3: non-poly(A) approach, mock sample versus non-poly(A)–
number of supporting factors and the percentage of circAtlas- selected (RNAs treated with depletion of both ribosomal RNAs and
specific circRNAs (Fig 8). This revealed the additive effects of polyadenylated RNAs) sample (two pairs).
these factors on circRNA reliability. Particularly, more than 70% of
circRNAs without any supporting factors were circAtlas-specific Of note, the downloaded mock-treated sample pairs should
circRNAs, whereas such a percentage sharply declined to ≤3% simultaneously satisfy the following criteria: (1) the number of
when considering the circRNAs with at least five supporting factors the circRNA candidates detected in the mock samples should
(Fig 8), supporting the effectiveness of these factors in determining be greater than 600 and (2) more than one-third of the circRNA
circRNA reliability before performing experimental validations. To candidates detected in the mock samples should be detected in the

Important indicators of circRNA reliability Chuang et al. https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 11 of 16

Figure 8. The correlation between number of supporting factors and the percentage of high-confidence circRNAs (RT-independent or RT-/non-RT–validated
circRNAs) and circAtlas-specific circRNAs for the 480,471 circAtlas circRNA candidates.
The right panel showed a clearer graph of the correlation between number of supporting factors and the percentage of circAtlas-specific circRNAs. The eight important
factors illustrated in Fig 6 except for the factor of “supporting BSJ read count” were considered because this factor was dependent on the examined samples and the
corresponding RNA-seq data. For the factors of “number of samples” and “number of supporting functional features,” three was used as a cutoff value. The detailed
information can be found in Table S3.
corresponding treated samples. For each RNA-seq dataset, dupli-
cated RNA-seq reads were removed using FastUniq (77).
Treat/mock ratio calculation
For accuracy, only the circRNAs detected in the mock samples with
the support of at least two BSJ reads were considered in the
analysis for each mock-treated sample pair. The expression levels
of circRNAs were calculated using the BSJ reads per million raw
reads (RPM). For each considered circRNA candidate, we calculated
the ratio of the circRNA expression detected in the treated samples
to that detected in the corresponding mock samples (i.e., treat/
mock ratio). The not-depleted and depleted circRNAs are defined as
follows:
(
≥ 1; Not − depleted circRNAs
treat=mock ratio
< 1; Depleted circRNAs:
Extractions of various factors
The factors of CIRI-full–identified full-length circular sequence,
number of circRNA detection tools, conservation patterns at the
individual (or sample), tissue, and species levels (including tissue
specificity index) were extracted from circAtlas 2.0 (22). The BSJ
events with the direct support of full-length circular sequences
from nanopore long RNA-seq reads were downloaded from Liu
et al’s study (44). The evolutionary rates were determined by phyloP
(45) or phastCons (46) scores, which were downloaded from the
UCSC Genome Browser at https://genome.ucsc.edu/. For each BSJ,
we considered the four regions around the BSJ: (1) within +1 to +10
nucleotides of the acceptor site; (2) within −10 to −1 nucleotides of
the acceptor site; (3) within −10 to −1 nucleotides of the donor site;
Important indicators of circRNA reliability Chuang et al.
and (4) within +1 to +10 nucleotides of the donor site. The evolu-
tionary rate of each region was measured by the average value of
the phyloP (or phastCons) scores of the considered nucleotides
(10 bp) within the region. The annotated exon boundaries and
alternative splicing (AS) patterns were derived from the Ensembl
annotation (version 100). The splice site strength of BSJs was es-
timated using MaxEntScan based on three scoring models (maxi-
mum entropy model, first-order Markov model, and weight matrix
model) (78). The numbers of reverse complementary sequences
(RCSs) across the flanking sequences (RCSacross; ±20,000 nucleo-
tides of the BSJs) or within individual flanking sequences (RCSwithin)
of the BSJs were calculated using CircMiMi (79) with the command
line: circmimi_tools check RCS –dist 20,000 genome.fa circRNAs.tsv
out.tsv. RBPs binding to the flanking regions (±1,000 nucleotides)
of BSJs was determined according to CLIP-supported RBP-binding
sites downloaded from ENCORI (80) at http://starbase.sysu.edu.cn/.
G-quadruplexes across the BSJs were determined using the QGRS
mapping algorithm with default parameters (81, 82). The predicted
G-quadruplexes should span the BSJ boundaries by ≥5 bp on both
sides of the BSJs. miRNA-binding sites across the BSJs were de-
termined using miRanda 3.3a (83) with pairing score ≥ 155. RBP-
binding sites across the BSJs were determined using RBPmap (84)
with a high stringency level and conservation filter. The predicted
miRNA (or RBP)-binding sites should span the BSJ boundaries
by ≥5 (or ≥2) bp on both sides of the BSJs, respectively. The seven
types of evidence for coding potential of more than 320,000
human circRNAs (including circAtlas circRNAs) were extracted
from TransCirc (85). The seven types of evidence included
ribosome/polysome-binding evidence, experimentally supported
translation initiation site, internal ribosome entry site, predicted
m6A modification site, circRNA-specific ORF, sequence composition
score for the ORF, and mass spectrometry data–supported peptide
across the BSJ. The translatable scores of circRNAs were down-
loaded from TransCirc.
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 12 of 16
Extractions of high-confidence circRNAs and circAtlas-specific
circRNAs
The high-confidence circRNAs examined here included RT-
independent and RT-/non-RT–validated circRNAs (see text). The
RT-independent circRNAs (1,478 events), which were supported by
both avian myeloblastosis virus- and Moloney murine leukemia
virus-derived RNA-seq reads, were extracted from our previous
study (8). The RT-/non-RT–validated circRNAs (553 events), in which
the NCL junctions (or their functions) were validated by RT-based
experiments and at least one type of non-RT–based experiments
in human tissues/cell lines, were extracted from CircR2disease
(version 2.0) (55). The circAtlas-specific circRNA candidates were the
BSJ events that were collected in circAtlas only rather than in eight
other publicly accessible circRNA databases (including CIRCpedia
v2 (86), CircRic (87), MiOncoCirc v2.0 (88), TSCD (51), circBase (89),
circRNADb (90), and exoRBase 2.0 (91)). All the circRNA candidates
deposited in CircR2disease or non-circAtlas circRNA databases
were not considered if the NCL junction coordinates were not
available or cannot be converted to the genomic coordinates on the
GRCh38 assembly by liftOver (92). For the NCL junctions with non-
assigned strands, the junction coordinates were matched to the
circRNA candidates in circAtlas. The analyzed circRNA candidates,
the related factors, and the treat/mock ratios of the corresponding
mock-treated sample pairs are given in Table S3. All the coordinates
in the table were one-based.
Statistics analysis
For enrichment analyses, we compared not-depleted circRNAs with
depleted ones in terms of various factors related to circRNA
identification, conservation, biogenesis, and function. P-values
were determined using two-tailed Fisher’s exact test or Wilcoxon
rank-sum test (WRST; greater and less) (see also the figure legends).
P-values were further FDR adjusted across 19 mock-treated sample
pairs for each examined feature using Benjamini–Hochberg cor-
rection. The Wilcoxon effect sizes and the corresponding 95%
confidence intervals were calculated with the wilcox_effsize
function implemented in the rstatix R package (93). To measure the
relative effect of individual factor in determining circRNA reliability,
we first employed a GLM with all the examined factors using the
statsmodels package (0.13.2) (94) in Python and then performed the
RCVE (50) as follows:
glm ðy ~ f 1 + f 2 + f 3 + f 4 + f 5 + f 6 + f 7 + f 8 ; family = binomialÞ
2
rall − rreduced
2
RCVE = 2 of exonic circular RNAs. Wiley Interdiscip Rev RNA 6: 563–579. doi:10.1002/
rall
In the GLM, y indicated whether a circRNA candidate was a not-
depleted circRNA (i.e., not-depleted circRNA = 1; depleted circRNA =
0). f1~f8 represented the eight factors examined in the mode:
supporting BSJ read count; whether BSJs were detected by multiple
tools; whether BSJs can be reconstructed full-length circRNAs;
number of samples observed the circRNAs; whether BSJs agreed to
annotated exon boundaries; whether BSJs had both donor and
acceptor splice sites matching annotated exon boundaries from the
Important indicators of circRNA reliability Chuang et al.
same colinear transcript isoforms; whether BSJs were subject to AS;
and number of supporting functional features, respectively. In the
2
RCVE formula, rall 2
and rreduced represented the r2 values calculated
by the GLM including all the eight factors and excluding the factor of
interest, respectively. The RCVE scores range from 0 to 1, with a
higher RCVE score indicating a more important contribution of the
examined factor to the model. The detailed results of the statistical
significance tests for each examined factor are given in Table S4.
Data Availability
The data underlying this study are available in the text and in
Tables S1–S4. Tables S1–S4 and all the codes used to generate the
results are available at GitHub (https://github.com/TreesLab/
circRNA_features).
Supplementary Information
Supplementary Information is available at https://doi.org/10.26508/lsa.
202201793.
Acknowledgements
This work was supported by T-J Chuang, Genomics Research Center, Aca-
demia Sinica, Taiwan and the National Science and Technology Council,
Taiwan (MOST 111-2311-B-001-021). We thank the related organizations for
generating and providing the RNA-seq data used in this study. We also thank
other members of T-J Chuang laboratory for helpful discussions.
Author Contributions
T-J Chuang: conceptualization, formal analysis, investigation,
methodology, and writing—original draft, review, and editing.
T-W Chiang and C-Y Chen: software and methodology.
Conflict of Interest Statement
The authors declare that they have no conflict of interest.
References
1 Chen I, Chen CY, Chuang TJ (2015) Biogenesis, identification, and function
wrna.1294
2 Chen LL (2020) The expanding regulatory mechanisms and cellular
functions of circular RNAs. Nat Rev Mol Cell Biol 21: 475–490. doi:10.1038/
s41580-020-0243-y
3 Jeck WR, Sorrentino JA, Wang K, Slevin MK, Burd CE, Liu J, Marzluff WF,
Sharpless NE (2013) Circular RNAs are abundant, conserved, and
associated with ALU repeats. RNA 19: 141–157. doi:10.1261/rna.035667.112
4 Salzman J, Chen RE, Olsen MN, Wang PL, Brown PO (2013) Cell-type
specific features of circular RNA expression. PLoS Genet 9: e1003777.
doi:10.1371/journal.pgen.1003777
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 13 of 16
 5 Guo JU, Agarwal V, Guo H, Bartel DP (2014) Expanded identification and
characterization of mammalian circular RNAs. Genome Biol 15: 409.
doi:10.1186/s13059-014-0409-z
6 Salzman J, Gawad C, Wang PL, Lacayo N, Brown PO (2012) Circular RNAs
are the predominant transcript isoform from hundreds of human genes
in diverse cell types. PLoS One 7: e30733. doi:10.1371/
journal.pone.0030733
7 Zheng Q, Bao C, Guo W, Li S, Chen J, Chen B, Luo Y, Lyu D, Li Y, Shi G, et al
(2016) Circular RNA profiling reveals an abundant circHIPK3 that
regulates cell growth by sponging multiple miRNAs. Nat Commun 7: 11215.
doi:10.1038/ncomms11215
8 Chuang TJ, Chen YJ, Chen CY, Mai TL, Wang YD, Yeh CS, Yang MY, Hsiao YT,
Chang TH, Kuo TC, et al (2018) Integrative transcriptome sequencing
reveals extensive alternative trans-splicing and cis-backsplicing in
human cells. Nucleic Acids Res 46: 3671–3691. doi:10.1093/nar/gky032
9 Chen LL, Yang L (2015) Regulation of circRNA biogenesis. RNA Biol 12:
381–388. doi:10.1080/15476286.2015.1020271
10 Yu CY, Liu HJ, Hung LY, Kuo HC, Chuang TJ (2014) Is an observed non-co-
linear RNA product spliced in trans, in cis or just in vitro? Nucleic Acids
Res 42: 9410–9423. doi:10.1093/nar/gku643
11 Rybak-Wolf A, Stottmeister C, Glazar P, Jens M, Pino N, Giusti S, Hanan M,
Behm M, Bartok O, Ashwal-Fluss R, et al (2015) Circular RNAs in the
mammalian brain are highly abundant, conserved, and dynamically
expressed. Mol Cell 58: 870–885. doi:10.1016/j.molcel.2015.03.027
12 Chen YJ, Chen CY, Mai TL, Chuang CF, Chen YC, Gupta SK, Yen L, Wang YD,
Chuang TJ (2020) Genome-wide, integrative analysis of circular RNA
dysregulation and the corresponding circular RNA-microRNA-mRNA
regulatory axes in autism. Genome Res 30: 375–391. doi:10.1101/
gr.255463.119
13 Huang A, Zheng H, Wu Z, Chen M, Huang Y (2020) Circular RNA-protein
interactions: Functions, mechanisms, and identification. Theranostics 10:
3503–3517. doi:10.7150/thno.42174
14 You X, Vlatkovic I, Babic A, Will T, Epstein I, Tushev G, Akbalik G, Wang M,
Glock C, Quedenau C, et al (2015) Neural circular RNAs are derived from
synaptic genes and regulated by development and plasticity. Nat
Neurosci 18: 603–610. doi:10.1038/nn.3975
15 Gasparini S, Licursi V, Presutti C, Mannironi C (2020) The secret garden of
neuronal circRNAs. Cells 9: 1815. doi:10.3390/cells9081815
16 Piwecka M, Glažar P, Hernandez-Miranda LR, Memczak S, Wolf SA, Rybak-
Wolf A, Filipchyk A, Klironomos F, Cerda Jara CA, Fenske P, et al (2017) Loss
of a mammalian circular RNA locus causes miRNA deregulation and
affects brain function. Science 357: eaam8526. doi:10.1126/
science.aam8526
17 Mai TL, Chen CY, Chen YC, Chiang TW, Chuang TJ (2022) Trans-genetic
effects of circular RNA expression quantitative trait loci and potential
causal mechanisms in autism. Mol Psychiatry 27: 4695–4706. doi:10.1038/
s41380-022-01714-4
18 Liu J, Zhang X, Yan M, Li H (2020) Emerging role of circular RNAs in cancer.
Front Oncol 10: 663. doi:10.3389/fonc.2020.00663
19 Rajappa A, Banerjee S, Sharma V, Khandelia P (2020) Circular RNAs:
Emerging role in cancer diagnostics and therapeutics. Front Mol Biosci 7:
577938. doi:10.3389/fmolb.2020.577938
20 Chen L, Wang C, Sun H, Wang J, Liang Y, Wang Y, Wong G (2021) The
bioinformatics toolbox for circRNA discovery and analysis. Brief
Bioinform 22: 1706–1728. doi:10.1093/bib/bbaa001
21 Xia S, Feng J, Chen K, Ma Y, Gong J, Cai F, Jin Y, Gao Y, Xia L, Chang H, et al
(2018) CSCD: A database for cancer-specific circular RNAs. Nucleic Acids
Res 46: D925–D929. doi:10.1093/nar/gkx863
22 Wu W, Ji P, Zhao F (2020) CircAtlas: An integrated resource of one million
highly accurate circular RNAs from 1070 vertebrate transcriptomes.
Genome Biol 21: 101. doi:10.1186/s13059-020-02018-y
Important indicators of circRNA reliability Chuang et al.
23 Vromman M, Vandesompele J, Volders PJ (2021) Closing the circle: Current
state and perspectives of circular RNA databases. Brief Bioinform 22:
288–297. doi:10.1093/bib/bbz175
24 Hansen TB, Veno MT, Damgaard CK, Kjems J (2016) Comparison of circular
RNA prediction tools. Nucleic Acids Res 44: e58. doi:10.1093/nar/gkv1458
25 Zeng X, Lin W, Guo M, Zou Q (2017) A comprehensive overview and
evaluation of circular RNA detection tools. PLoS Comput Biol 13:
e1005420. doi:10.1371/journal.pcbi.1005420
26 Chen CY, Chuang TJ (2019) Comment on [L8D2Q2M0]A comprehensive
overview and evaluation of circular RNA detection tools[R8D2Q2M1].
PLoS Comput Biol 15: e1006158. doi:10.1371/journal.pcbi.1006158
27 Szabo L, Salzman J (2016) Detecting circular RNAs: Bioinformatic and
experimental challenges. Nat Rev Genet 17: 679–692. doi:10.1038/
nrg.2016.114
28 Memczak S, Jens M, Elefsinioti A, Torti F, Krueger J, Rybak A, Maier L,
Mackowiak SD, Gregersen LH, Munschauer M, et al (2013) Circular RNAs
are a large class of animal RNAs with regulatory potency. Nature 495:
333–338. doi:10.1038/nature11928
29 Zhang XO, Wang HB, Zhang Y, Lu X, Chen LL, Yang L (2014) Complementary
sequence-mediated exon circularization. Cell 159: 134–147. doi:10.1016/
j.cell.2014.09.001
30 Zhang XO, Dong R, Zhang Y, Zhang JL, Luo Z, Zhang J, Chen LL, Yang L (2016)
Diverse alternative back-splicing and alternative splicing landscape of
circular RNAs. Genome Res 26: 1277–1287. doi:10.1101/gr.202895.115
31 Jakobi T, Uvarovskii A, Dieterich C (2019) circtools-a one-stop software
solution for circular RNA research. Bioinformatics 35: 2326–2328.
doi:10.1093/bioinformatics/bty948
32 Xiao MS, Wilusz JE (2019) An improved method for circular RNA
purification using RNase R that efficiently removes linear RNAs
containing G-quadruplexes or structured 3’ ends. Nucleic Acids Res 47:
8755–8769. doi:10.1093/nar/gkz576
33 Zhang Y, Yang L, Chen LL (2016) Characterization of circular RNAs.
Methods Mol Biol 1402: 215–227. doi:10.1007/978-1-4939-3378-5_17
34 Rabin A, Zaffagni M, Ashwal-Fluss R, Patop IL, Jajoo A, Shenzis S, Carmel L,
Kadener S (2021) SRCP: A comprehensive pipeline for accurate
annotation and quantification of circRNAs. Genome Biol 22: 277.
doi:10.1186/s13059-021-02497-7
35 Chen CY, Chuang TJ (2019) NCLcomparator: Systematically post-
screening non-co-linear transcripts (circular, trans-spliced, or fusion
RNAs) identified from various detectors. BMC Bioinformatics 20: 3.
doi:10.1186/s12859-018-2589-0
36 Petkovic S, Muller S (2015) RNA circularization strategies in vivo and in
vitro. Nucleic Acids Res 43: 2454–2465. doi:10.1093/nar/gkv045
37 Hansen TB (2018) Improved circRNA identification by combining
prediction algorithms. Front Cell Dev Biol 6: 20. doi:10.3389/
fcell.2018.00020
38 Gao Y, Wang J, Zhao F (2015) CIRI: An efficient and unbiased algorithm for
de novo circular RNA identification. Genome Biol 16: 4. doi:10.1186/
s13059-014-0571-3
39 Gao Y, Wang J, Zheng Y, Zhang J, Chen S, Zhao F (2016) Comprehensive
identification of internal structure and alternative splicing events in
circular RNAs. Nat Commun 7: 12060. doi:10.1038/ncomms12060
40 Mercer TR, Clark MB, Andersen SB, Brunck ME, Haerty W, Crawford J, Taft
RJ, Nielsen LK, Dinger ME, Mattick JS (2015) Genome-wide discovery of
human splicing branchpoints. Genome Res 25: 290–303. doi:10.1101/
gr.182899.114
41 Li Y, Wang F, Teng P, Ku L, Chen L, Feng Y, Yao B (2022) Accurate
identification of circRNA landscape and complexity reveals their pivotal
roles in human oligodendroglia differentiation. Genome Biol 23: 48.
doi:10.1186/s13059-022-02621-1
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 14 of 16
42 Davis CA, Hitz BC, Sloan CA, Chan ET, Davidson JM, Gabdank I, Hilton JA,
Jain K, Baymuradov UK, Narayanan AK, et al (2018) The encyclopedia of
DNA elements (ENCODE): Data portal update. Nucleic Acids Res 46:
D794–D801. doi:10.1093/nar/gkx1081
43 Zheng Y, Ji P, Chen S, Hou L, Zhao F (2019) Reconstruction of full-length
circular RNAs enables isoform-level quantification. Genome Med 11: 2.
doi:10.1186/s13073-019-0614-1
44 Liu Z, Tao C, Li S, Du M, Bai Y, Hu X, Li Y, Chen J, Yang E (2021) circFL-seq
reveals full-length circular RNAs with rolling circular reverse
transcription and nanopore sequencing. Elife 10: e69457. doi:10.7554/
elife.69457
45 Pertea M, Pertea GM, Salzberg SL (2011) Detection of lineage-specific
evolutionary changes among primate species. BMC Bioinformatics 12:
274. doi:10.1186/1471-2105-12-274
46 Siepel A, Bejerano G, Pedersen JS, Hinrichs AS, Hou M, Rosenbloom K,
Clawson H, Spieth J, Hillier LW, Richards S, et al (2005) Evolutionarily
conserved elements in vertebrate, insect, worm, and yeast genomes.
Genome Res 15: 1034–1050. doi:10.1101/gr.3715005
47 Ashwal-Fluss R, Meyer M, Pamudurti NR, Ivanov A, Bartok O, Hanan M,
Evantal N, Memczak S, Rajewsky N, Kadener S (2014) circRNA biogenesis
competes with pre-mRNA splicing. Mol Cell 56: 55–66. doi:10.1016/
j.molcel.2014.08.019
48 Starke S, Jost I, Rossbach O, Schneider T, Schreiner S, Hung LH, Bindereif
A (2015) Exon circularization requires canonical splice signals. Cell Rep
10: 103–111. doi:10.1016/j.celrep.2014.12.002
49 Wang Y, Wang Z (2015) Efficient backsplicing produces translatable
circular mRNAs. RNA 21: 172–179. doi:10.1261/rna.048272.114
50 Kvikstad EM, Tyekucheva S, Chiaromonte F, Makova KD (2007) A macaque
[R8S2Q1M7]s-eye view of human insertions and deletions: Differences in
mechanisms. PLoS Comput Biol 3: 1772–1782. doi:10.1371/
journal.pcbi.0030176
51 Xia S, Feng J, Lei L, Hu J, Xia L, Wang J, Xiang Y, Liu L, Zhong S, Han L, et al
(2017) Comprehensive characterization of tissue-specific circular RNAs
in the human and mouse genomes. Brief Bioinform 18: 984–992.
doi:10.1093/bib/bbw081
52 Houseley J, Tollervey D (2010) Apparent non-canonical trans-splicing is
generated by reverse transcriptase in vitro. PLoS One 5: e12271.
doi:10.1371/journal.pone.0012271
53 Wu CS, Yu CY, Chuang CY, Hsiao M, Kao CF, Kuo HC, Chuang TJ (2014)
Integrative transcriptome sequencing identifies trans-splicing events
with important roles in human embryonic stem cell pluripotency.
Genome Res 24: 25–36. doi:10.1101/gr.159483.113
54 McManus CJ, Duff MO, Eipper-Mains J, Graveley BR (2010) Global analysis
of trans-splicing in Drosophila. Proc Natl Acad Sci U S A 107: 12975–12979.
doi:10.1073/pnas.1007586107
55 Fan C, Lei X, Tie J, Zhang Y, Wu FX, Pan Y (2022) CircR2Disease v2.0: An
updated web server for experimentally validated circRNA-disease
associations and its application. Genomics Proteomics Bioinformatics
20: 435–445. doi:10.1016/j.gpb.2021.10.002
56 Westholm JO, Miura P, Olson S, Shenker S, Joseph B, Sanfilippo P, Celniker
SE, Graveley BR, Lai EC (2014) Genome-wide analysis of drosophila
circular RNAs reveals their structural and sequence properties and age-
dependent neural accumulation. Cell Rep 9: 1966–1980. doi:10.1016/
j.celrep.2014.10.062
57 Hoffmann S, Otto C, Doose G, Tanzer A, Langenberger D, Christ S, Kunz M,
Holdt LM, Teupser D, Hackermuller J, et al (2014) A multi-split mapping
algorithm for circular RNA, splicing, trans-splicing and fusion detection.
Genome Biol 15: R34. doi:10.1186/gb-2014-15-2-r34
58 Gaffo E, Buratin A, Dal Molin A, Bortoluzzi S (2022) Sensitive, reliable and
robust circRNA detection from RNA-seq with CirComPara2. Brief
Bioinform 23: bbab418. doi:10.1093/bib/bbab418
Important indicators of circRNA reliability Chuang et al.
59 Kim P, Yoon S, Kim N, Lee S, Ko M, Lee H, Kang H, Kim J, Lee S (2010)
ChimerDB 2.0–A knowledgebase for fusion genes updated. Nucleic Acids
Res 38: D81–D85. doi:10.1093/nar/gkp982
60 Al-Balool HH, Weber D, Liu Y, Wade M, Guleria K, Nam PLP, Clayton J, Rowe
W, Coxhead J, Irving J, et al (2011) Post-transcriptional exon shuffling
events in humans can be evolutionarily conserved and abundant.
Genome Res 21: 1788–1799. doi:10.1101/gr.116442.110
61 Chuang TJ, Wu CS, Chen CY, Hung LY, Chiang TW, Yang MY (2016) NCLscan:
Accurate identification of non-co-linear transcripts (fusion, trans-
splicing and circular RNA) with a good balance between sensitivity and
precision. Nucleic Acids Res 44: e29. doi:10.1093/nar/gkv1013
62 Takahara T, Kanazu SI, Yanagisawa S, Akanuma H (2000) Heterogeneous
Sp1 mRNAs in human HepG2 cells include a product of homotypic trans-
splicing. J Biol Chem 275: 38067–38072. doi:10.1074/jbc.m002010200
63 Flouriot G, Brand H, Seraphin B, Gannon F (2002) Natural trans-spliced
mRNAs are generated from the human estrogen receptor-alpha (hER
alpha) gene. J Biol Chem 277: 26244–26251. doi:10.1074/jbc.m203513200
64 Horiuchi T, Aigaki T (2006) Alternative trans-splicing: A novel mode of
pre-mRNA processing. Biol Cell 98: 135–140. doi:10.1042/bc20050002
65 Gingeras TR (2009) Implications of chimaeric non-co-linear transcripts.
Nature 461: 206–211. doi:10.1038/nature08452
66 Kelly S, Greenman C, Cook PR, Papantonis A (2015) Exon skipping is
correlated with exon circularization. J Mol Biol 427: 2414–2417.
doi:10.1016/j.jmb.2015.02.018
67 Zhang P, Zhang XO, Jiang T, Cai L, Huang X, Liu Q, Li D, Lu A, Liu Y, Xue W,
et al (2020) Comprehensive identification of alternative back-splicing in
human tissue transcriptomes. Nucleic Acids Res 48: 1779–1789.
doi:10.1093/nar/gkaa005
68 Lu T, Cui L, Zhou Y, Zhu C, Fan D, Gong H, Zhao Q, Zhou C, Zhao Y, Lu D, et al
(2015) Transcriptome-wide investigation of circular RNAs in rice. RNA 21:
2076–2087. doi:10.1261/rna.052282.115
69 Errichelli L, Dini Modigliani S, Laneve P, Colantoni A, Legnini I, Capauto D,
Rosa A, De Santis R, Scarfo R, Peruzzi G, et al (2017) FUS affects circular
RNA expression in murine embryonic stem cell-derived motor neurons.
Nat Commun 8: 14741. doi:10.1038/ncomms14741
70 Shen H, An O, Ren X, Song Y, Tang SJ, Ke XY, Han J, Tay DJT, Ng VHE, Molias
FB, et al (2022) ADARs act as potent regulators of circular transcriptome
in cancer. Nat Commun 13: 1508. doi:10.1038/s41467-022-29138-2
71 Yang L, Wilusz JE, Chen LL (2022) Biogenesis and regulatory roles of
circular RNAs. Annu Rev Cell Dev Biol 38: 263–289. doi:10.1146/annurev-
cellbio-120420-125117
72 Gao Y, Zhang J, Zhao F (2018) Circular RNA identification based on
multiple seed matching. Brief Bioinform 19: 803–810. doi:10.1093/bib/
bbx014
73 Dong R, Ma XK, Chen LL, Yang L (2019) Genome-Wide annotation of
circRNAs and their alternative back-splicing/splicing with CIRCexplorer
pipeline. Methods Mol Biol 1870: 137–149. doi:10.1007/978-1-4939-8808-
2_10
74 Cheng J, Metge F, Dieterich C (2016) Specific identification and
quantification of circular RNAs from sequencing data. Bioinformatics 32:
1094–1096. doi:10.1093/bioinformatics/btv656
75 Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P,
Chaisson M, Gingeras TR (2013) STAR: Ultrafast universal RNA-seq aligner.
Bioinformatics 29: 15–21. doi:10.1093/bioinformatics/bts635
76 Li H, Durbin R (2009) Fast and accurate short read alignment with
Burrows-Wheeler transform. Bioinformatics 25: 1754–1760. doi:10.1093/
bioinformatics/btp324
77 Xu H, Luo X, Qian J, Pang X, Song J, Qian G, Chen J, Chen S (2012) FastUniq: A
fast de novo duplicates removal tool for paired short reads. PLoS One 7:
e52249. doi:10.1371/journal.pone.0052249
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 15 of 16
78 Yeo G, Burge CB (2004) Maximum entropy modeling of short sequence
motifs with applications to RNA splicing signals. J Comput Biol 11:
377–394. doi:10.1089/1066527041410418
79 Chiang TW, Mai TL, Chuang TJ (2022) CircMiMi: A stand-alone software for
constructing circular RNA-microRNA-mRNA interactions across species.
BMC Bioinformatics 23: 164. doi:10.1186/s12859-022-04692-0
80 Li JH, Liu S, Zhou H, Qu LH, Yang JH (2014) starBase v2.0: decoding miRNA-
ceRNA, miRNA-ncRNA and protein-RNA interaction networks from large-
scale CLIP-Seq data. Nucleic Acids Res 42: D92–D97. doi:10.1093/nar/
gkt1248
81 Frees S, Menendez C, Crum M, Bagga PS (2014) QGRS-conserve: A
computational method for discovering evolutionarily conserved G-
quadruplex motifs. Hum Genomics 8: 8. doi:10.1186/1479-7364-8-8
82 Menendez C, Frees S, Bagga PS (2012) QGRS-H predictor: A web server for
predicting homologous quadruplex forming G-rich sequence motifs in
nucleotide sequences. Nucleic Acids Res 40: W96–W103. doi:10.1093/nar/
gks422
83 Enright AJ, John B, Gaul U, Tuschl T, Sander C, Marks DS (2003) MicroRNA
targets in Drosophila. Genome Biol 5: R1. doi:10.1186/gb-2003-5-1-r1
84 Paz I, Kosti I, Ares M Jr., Cline M, Mandel-Gutfreund Y (2014) RBPmap: A
web server for mapping binding sites of RNA-binding proteins. Nucleic
Acids Res 42: W361–W367. doi:10.1093/nar/gku406
85 Huang W, Ling Y, Zhang S, Xia Q, Cao R, Fan X, Fang Z, Wang Z, Zhang G
(2021) TransCirc: An interactive database for translatable circular RNAs
based on multi-omics evidence. Nucleic Acids Res 49: D236–D242.
doi:10.1093/nar/gkaa823
86 Dong R, Ma XK, Li GW, Yang L (2018) CIRCpedia v2: An updated database
for comprehensive circular RNA annotation and expression comparison.
Genomics Proteomics Bioinformatics 16: 226–233. doi:10.1016/
j.gpb.2018.08.001
Important indicators of circRNA reliability Chuang et al.
87 Ruan H, Xiang Y, Ko J, Li S, Jing Y, Zhu X, Ye Y, Zhang Z, Mills T, Feng J, et al
(2019) Comprehensive characterization of circular RNAs in ~ 1000 human
cancer cell lines. Genome Med 11: 55. doi:10.1186/s13073-019-0663-5
88 Vo JN, Cieslik M, Zhang Y, Shukla S, Xiao L, Zhang Y, Wu YM, Dhanasekaran
SM, Engelke CG, Cao X, et al (2019) The landscape of circular RNA in
cancer. Cell 176: 869–881.e13. doi:10.1016/j.cell.2018.12.021
89 Glazar P, Papavasileiou P, Rajewsky N (2014) circBase: a database for
circular RNAs. RNA 20: 1666–1670. doi:10.1261/rna.043687.113
90 Chen X, Han P, Zhou T, Guo X, Song X, Li Y (2016) circRNADb: A
comprehensive database for human circular RNAs with protein-coding
annotations. Sci Rep 6: 34985. doi:10.1038/srep34985
91 Lai H, Li Y, Zhang H, Hu J, Liao J, Su Y, Li Q, Chen B, Li C, Wang Z, et al (2022)
exoRBase 2.0: An atlas of mRNA, lncRNA and circRNA in extracellular
vesicles from human biofluids. Nucleic Acids Res 50: D118–D128.
doi:10.1093/nar/gkab1085
92 Hinrichs AS, Karolchik D, Baertsch R, Barber GP, Bejerano G, Clawson H,
Diekhans M, Furey TS, Harte RA, Hsu F, et al (2006) The UCSC genome
browser database: Update 2006. Nucleic Acids Res 34: D590–D598.
doi:10.1093/nar/gkj144
93 Tomczak M, Tomczak E (2014) The need to report effect size estimates
revisited. An overview of some recommended measures of effect size.
Trends Sport Sci 1: 19–25.
94 Seabold S, Perktold J (2010) statsmodels: Econometric and statistical
modeling with python. In Proceedings of the 9th Python in Science
Conference. Austin, TX: SciPy Organizers. doi:10.25080/Majora-92bf1922-011
License: This article is available under a Creative
Commons License (Attribution 4.0 International, as
described at https://creativecommons.org/
licenses/by/4.0/).
https://doi.org/10.26508/lsa.202201793 vol 6 | no 5 | e202201793 16 of 16