![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
|
Genome Res. 17:852-864, 2007 ©2007 by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/07 $5.00 OPEN ACCESS ARTICLE Letter Structured RNAs in the ENCODE selected regions of the human genome1 Institute for Theoretical Chemistry, University of Vienna, A-1090 Wien, Austria; 2 Center for Biomolecular Science and Engineering, University of California, Santa Cruz, Santa Cruz, California 95064, USA; 3 Molecular Biophysics and Biochemistry Department, Yale University, New Haven, Connecticut 06520-8114, USA; 4 European Molecular Biology Laboratory, 69117 Heidelberg, Germany; 5 Bioinformatics Group, Department of Computer Science, University of Leipzig, D-04107 Leipzig, Germany; 6 Fraunhofer Institute for Cell Therapy and Immunology, 04103 Leipzig, Germany; 7 Grup de Recerca en Informática Biomèdica, Institut Municipal d’Investigació Mèdica/Universitat Pompeu Fabra. Passeig Marítim de la Barceloneta, 37-49,08003, Barcelona, Catalonia, Spain; 8 Affymetrix, Inc., Santa Clara, California 95051, USA; 9 Center for Integrative Genomics, University of Lausanne, 1015 Lausanne, Switzerland; 10 Department of Genetic Medicine and Development, University of Geneva Medical School, 1211 Geneva, Switzerland; 11 Molecular, Cellular and Developmental Biology Department, Yale University, New Haven, Connecticut 06520-8114, USA; 12 Santa Fe Institute, Santa Fe, New Mexico 87501 USA; 13 Department of Ecology and Evolutionary Biology; Yale University, New Haven, CT 06520-8106, USA
Functional RNA structures play an important role both in the context of noncoding RNA transcripts as well as regulatory elements in mRNAs. Here we present a computational study to detect functional RNA structures within the ENCODE regions of the human genome. Since structural RNAs in general lack characteristic signals in primary sequence, comparative approaches evaluating evolutionary conservation of structures are most promising. We have used three recently introduced programs based on either phylogenetic–stochastic context-free grammar (EvoFold) or energy directed folding (RNAz and AlifoldZ), yielding several thousand candidate structures (corresponding to  2.7% of the ENCODE regions). EvoFold has its highest sensitivity in highly conserved and relatively AU-rich regions, while RNAz favors slightly GC-rich regions, resulting in a relatively small overlap between methods. Comparison with the GENCODE annotation points to functional RNAs in all genomic contexts, with a slightly increased density in 3'-UTRs. While we estimate a significant false discovery rate of 50%–70% many of the predictions can be further substantiated by additional criteria: 248 loci are predicted by both RNAz and EvoFold, and an additional 239 RNAz or EvoFold predictions are supported by the (more stringent) AlifoldZ algorithm. Five hundred seventy RNAz structure predictions fall into regions that show signs of selection pressure also on the sequence level (i.e., conserved elements). More than 700 predictions overlap with noncoding transcripts detected by oligonucleotide tiling arrays. One hundred seventy-five selected candidates were tested by RT-PCR in six tissues, and expression could be verified in 43 cases (24.6%).
The goal of The ENCODE Project Consortium (Encyclopedia of DNA Elements [ENCODE]) is the comprehensive analysis of functional elements in the human genome. One of its main goals is the thorough annotation of transcripts in terms of structure and function. Both genome-wide studies (Bertone et al. 2004  ; Carninci et al. 2005; Cheng et al. 2005) and the far more detailed studies targeted to the ENCODE regions (The ENCODE Project Consortium 2007) show a much more extensive and complex transcriptional map than previously anticipated, comprising a mosaic of overlapping transcription, antisense transcripts, abundant alternative splicing, and a plethora of novel transcribed elements. Using a series of sensitive methods, it was demonstrated that 93% of the ENCODE regions exist in primary nuclear transcripts in at least one of the tested tissues. An as-yet not satisfactorily resolved question is whether novel transcripts lacking protein-coding capacity (noncoding transcripts) have biological function as such, or whether they rather represent "biological noise" (i.e., selectively neutral transcription). Analogous to the analysis of protein-coding genes, a combination of both experimental and computational techniques seems necessary to address this question.
On the experimental side, we can draw upon the evidence from large-scale oligonucleotide tiling array studies performed on the ENCODE regions as well as a small set of verification experiments (The ENCODE Project Consortium 2007). Unfortunately, there is at present no general way to predict noncoding transcripts in eukaryotic genomes. A few methods exploit weak statistical signals like mutational strand bias, strand-specific selection against polyadenylation signals, or exclusion of repeat elements to predict transcribed regions in the genome (Semon and Duret 2004
RNA secondary structures are known to play an important functional role not only in noncoding transcripts, but also in the context of protein-coding mRNAs. Structural motifs serve regulatory functions in untranslated regions (Mignone et al. 2002 The comprehensive knowledge of encoded secondary structures in the genome is important to determine at which level DNA is actually functional, and without it, an "encyclopedia" of functional elements would be incomplete.
In this study, we use different comparative approaches to predict functional RNA secondary structures and provide a detailed comparison with the results of other ENCODE subprojects; in particular, experimental data from oligonucleotide tiling array studies. The computational approach is based on predicting consensus structures and the observation that structural constraints imply specific mutational patterns visible at the sequence level. EvoFold (Pedersen et al. 2006
Three approaches Almost all RNA molecules form secondary structures. The challenge is thus to recognize those sections of the genome in which the structure is more conserved than one would expect from primary sequence conservation alone. We employ here three fairly different methods that are designed to recognize evolutionarily conserved secondary structures. All three are based on given multiple-sequence alignments and attempt to (1) predict a consensus secondary structure for aligned sequences, and then (2) apply a test of whether or not the consensus structure found is unusual.
Consensus structures can be inferred either by means of energy-directed folding or using a phylo-SCFG model. The RNAalifold algorithm computes the most stable secondary structure that is compatible with the input alignment (Hofacker et al. 2002
AlifoldZ uses a random shuffle approach to estimate the expected background distribution (Washietl and Hofacker 2004
These limitations are overcome by RNAz (Washietl et al. 2005b
EvoFold is based on two competing phylo-SCFG models of RNA sequence evolution: a structural model, similar to the Pfold model, and a nonstructural model (Pedersen et al. 2006
Screening multispecies alignments of the ENCODE regions
For AlifoldZ, we used a sample of a maximum of 10 sequences from the alignments. The consensus minimum free energy (MFE) quantifying the stability of the consensus fold predicted by RNAalifold of all scanned windows is shown in Figure 1. This shows that some sort of consensus fold can be found in almost all alignments. It is not possible to discriminate on the basis of this score; therefore, the Z-score is calculated to assess its significance. We only considered Z-scores for alignments with consensus MFE <–15, since Z-scores can be unstable for low levels of consensus MFE. This filter is the most stringent one and leaves us with 660 and 348 hits, respectively, for the two significance cutoffs Z < –3.5 and Z < –4, which have been used by Washietl and Hofacker (2004)
In the case of the RNAz screen, we selected up to six sequences; if there were more than 10 sequences in the alignment, we selected three different samples of six. These were classified using the SVM. The SVM score distributions can be seen in Figure 1. For convenience, the SVM scores are converted to "RNA class probabilities," and we used two cutoffs, 0.5 and 0.9, as introduced by Washietl et al. (2005b) . This results in 7093 and 3707 predictions, respectively. All sequences of the alignments were used for EvoFold. First the regions were screened in fixed-size windows, then the predicted substructures were rescored and filtered for spurious predictions (short predictions with <10 base pairs [bp] were discarded). Based on the EvoFold score, we defined two sets: one with all predicted structures and one with the top 50% high-scoring structures, consisting of 9953 and 4986 predictions, respectively. From the score distributions in Figure 1 and the results in Table 1, one can see that all three methods apply a relatively stringent filter on the data: On the high-significance level, RNAz and EvoFold predict 1.4% and 1.3% of the ENCODE regions to form structural RNAs, which is in both cases <5% of the scored input alignments. Note that the input varies between RNAz and EvoFold because specific schemes were used to filter the raw alignments (for details, see Methods).
Estimating background signal An important issue in any genome-wide screen, be it experimental or computational, is the estimation of the false discovery rate. To this end, we repeated the analysis with randomly shuffled alignments (see Methods). This procedure is designed to remove correlations arising from secondary structures while leaving other characteristics of the aligned sequences untouched. Score distributions for the randomized data are shown in Figure 1, and the results of the randomized screens are summarized in Table 1.
An important aspect in the context of randomizing RNA secondary structures is dinucleotide content (Workman and Krogh 1999
We observe a relatively high false discovery rate for both RNAz and EvoFold (Table 2). On the highly significant set, the false discovery rate (after dinucleotide correction) is 50.0% for RNAz and 70.9% for EvoFold, respectively. Since the shuffling approach comes with uncertainties (Washietl and Hofacker 2004
Comparison of different predictions Figure 2 shows the overlap between different methods. Of the AlifoldZ hits, 70.9% overlap with the RNAz predictions. Since false positives are estimated to be at least 20% in AlifoldZ and false positives for RNAz and AlifoldZ arise for different reasons, this overlap is what can be expected. The 247 overlapping hits thus can be regarded as predictions with very high confidence. On the other hand, due to the very restrictive consensus MFE and Z-score cutoff used for AlifoldZ, many true RNAz hits will not yield an AlifoldZ signal.
The overlap between RNAz and EvoFold is extremely low. Only 7.2% of the RNAz hits overlap with EvoFold predictions. While this constitutes a 1.6-fold enrichment over the randomly expected overlap, and although the high estimated false discovery rates limit the best possible overlap to about one-third, this small overlap was unexpected. Close inspection of the data, however, revealed the interesting fact that RNAz and EvoFold essentially detect complementary RNA structures: While RNAz is sensitive to alignments with moderate and high GC content and relatively low sequence similarity, EvoFold has its peak sensitivities for low GC content and high sequence similarity (Fig. 3). Both methods were trained on structurally diverse subsets of the Rfam database with average GC contents of 50%. However, the parametrization of EvoFold’s nonstructural submodel creates a bias in its structural predictions toward AT-rich regions. The human genome has an overall GC content of 42%. Many of the known structured RNAs, such as microRNAs and H/ACA box snoRNAs, have an average GC content close to 50%; however, some have a relatively low GC content, such as tRNAs, which have an average GC content of 34%.
The second clear difference is that a large fraction of EvoFold predictions are within highly conserved alignments, while RNAz predictions essentially follow the conservation distribution found in the input regions. EvoFold, as opposed to RNAz, explicitly models the rate of substitution and was trained to detect slowly evolving RNA structures. Since many known ncRNAs are highly conserved not only in structure but also in sequence, this part of the conservation spectrum is of particular interest. However, due to the lack of sequence variation in these alignments, discriminating between true- and false-positive predictions is difficult. EvoFold is more sensitive for highly conserved alignments than RNAz, at the expense of a higher rate of false positives.
Detection of known ncRNAs
An interesting example is H19, which shows that long spliced transcripts can have structural "domains" and that structural ncRNAs are not necessarily small RNAs with a global structure as seen for tRNAs or snoRNAs. In addition to these well-described examples, we found seven overlapping EvoFold/RNAz hits with significant sequence similarity (BLAST E < 10–6) to the set of putative ncRNAs from the mouse Fantom2 project (Okazaki et al. 2002 ), supporting the role of these transcripts as functional ncRNAs.
Comparison with other ENCODE data
Of the high-significance RNAz hits, 22.3% overlap with experimentally detected sites of transcription. This includes UTR elements and the predictions in coding regions (see below). Without these regions (i.e., counting only intergenic and intronic), 15.7% of the RNAz hits overlap with TARs/Transfrags. This corresponds to a significant enrichment of approximately twofold. However, this must be interpreted with caution since TARs/Transfrags are very GC rich (unannotated Transfrags: 56%). It is unclear to what extent this bias has biological reasons or is the result of the hybridization technique, and consequently, it is difficult to interpret the significance of these enriched overlaps. GC content seems to be an important issue since we do not see any enrichment but, in fact, a small negative correlation of EvoFold hits and TARs/Transfrags (only 5.8% of the intergenic and intronic EvoFold hits overlap TARs/Transfrags). The sensitivity of tiling arrays on AU-rich sequences may be lower than for GC-rich sequences.
Another important issue in this context is that it is unclear how secondary structure affects detection performance on tiling arrays. Similar to previous studies (Clote et al. 2005 While much of the ENCODE region is alignable at least with the genomic DNA of closely related species, and hence used as input in the computational screens detailed above, only a subset of these sequences is under stabilizing selection at the sequence level. We therefore compared the structured RNA candidates with the multiple species analysis for sequence-constrained elements. We used the "moderate" set of constrained elements, which comprises regions detected by at least two of three conservation programs in at least two of three alignments prepared by different methods (Margulies et al. 2007). These conserved elements cover 4.9% of the ENCODE regions.
Eight hundred forty-one RNAz hits (22.69%) overlap with conserved regions, 570 (17.2%) without hits in UTRs and coding regions. For EvoFold predictions the overlap is much higher, 3579 (71.78%) including exons and 2130 (60.41%) without exons, in line with the program’s general tendency to predict structures in highly conserved regions. The fact that a large fraction of predicted conserved RNA structures does not correlate with high sequence conservation does not come as a surprise. Indeed, Torarinsson and colleagues reported expressed noncoding RNAs in regions that are not alignable between human and mouse and nevertheless have conserved secondary structures (Torarinsson et al. 2006 It seems noteworthy that all but one16 of the few known ncRNAs in the ENCODE regions overlap with constrained elements and TARs/Transfrags. This might be special for this set of snoRNAs and miRNAs, which are presumably abundantly expressed as kind of "housekeeping ncRNAs" and have well known reasons for sequence constraints. The 114 and 142 intergenic/intronic RNAz and EvoFold hits, respectively, that overlap both conserved elements and TARs/Transfrags are of special interest. Twenty-one of these are detected by both EvoFold and RNAz, while 12 of these have an AlifoldZ Z-score <–3.5. These numbers demonstrate that there is only a relatively small, but nonnegligible, number of structured ncRNAs that are similar to the "classical" ncRNA families in terms of high sequence conservation, highly stabilized and well-conserved secondary structures, and high expression levels.
Overlap with GENCODE annotations
RNAz predictions are depleted in coding regions despite the high GC content (53%). This is in keeping with the expectation that functional ncRNAs in coding regions should be rare. However, functional RNA structures do occur within coding regions, and thus these predictions are also of interest. As mentioned above, there are a few well-known functions assigned to hairpin structures within coding regions. In addition, there is recent evidence that secondary structures are much more widespread in coding regions of both prokaryotes (Katz and Burge 2003 ) and eukaryotes (Chamary and Hurst 2005; Meyer and Miklós 2005) than previously thought. EvoFold predictions are highly enriched in coding regions. However, the method has previously been shown to have above average rates of false positives in coding regions (Pedersen et al. 2006), presumably because of the high level of sequence conservation. The interpretation of these coding predictions is thus challenging and often requires additional evidence, such as conservation of synonymous codon positions (Pedersen et al. 2004a, b) or overlapping predictions from several methods. There are 41 overlapping RNAz/EvoFold hits from the high-significance sets in coding exons, 18 of which are particularly stable with AlifoldZ scores Z < –3.5. In general, we do not see any trend of noncoding structures favoring intronic over intergenic fractions. For RNAz, however, one can observe that "proximal" intergenic and intronic fractions are slightly enriched while distal fractions are depleted, i.e., we see more structures near genes and exons. For EvoFold, both intergenic and intronic fractions are depleted in favor of the more conserved UTR and coding regions. An interesting result of the GENCODE annotation project is the transcriptional complexity of protein-coding gene loci. For the 487 loci in the ENCODE regions, 2608 different transcripts were identified, 1511 of them noncoding. Two hundred twenty-nine and 940 RNAz and EvoFold hits, respectively, overlap with a noncoding GENCODE transcript. Some of these transcripts are extensively structured (Fig. 7F,G, see below).
Experimental verification of selected predictions
Examples of selected predictions Figure 7 shows some examples of predicted RNAs in different genomic contexts. A series of criteria supports the prediction of these regions as functional RNA: (1) several independent RNAz and/or EvoFold hits in close vicinity; (2) overlapping hits of EvoFold/RNAz; (3) additional support from AlifoldZ; (4) support from compensatory/consistent mutations in the predicted structures; (5) overlap with predictions of sequence constrained elements. Evidence for transcription of these regions comes from TARs/Transfrags, ESTs, or GENCODE transcripts (Harrow et al. 2006 ). In addition, we have performed 5'-RACE/microarray experiments (see Methods).
Examples A, B, and C (in Fig. 7A–C) are located within intergenic regions, all of them >50 kb away from any GENCODE annotation. There are also no "putative" or "pseudogene" GENCODE annotations or any predicted protein-coding genes close by. Nevertheless, we observe sequence-constrained elements. In all cases, the sequences are conserved across eutherian mammals, B is also conserved in chicken, and in C there is a sequence from opossum. We observe several RNAz and EvoFold hits in these regions. In A, for example, we have two independent RNAz hits, one overlapping with an EvoFold hit. This example illustrates the different "sweet spots" of the two programs. The significant RNAz hit is in the region of moderate conservation, while the overlapping hit with EvoFold within the highly conserved region is only of borderline significance. In all three examples there is additional support from AlifoldZ, which is particularly impressive for B and C with Z-scores of –9.5 and –7.0. We want to recall that this Z-score means standard deviations from the expected random background score for a given alignment. The transcription of these RNAs was confirmed by 5'-RACE/array analysis. Examples D and E (in Fig. 7D,E) show two sequence-constrained "islands" in introns of well-known protein-coding genes. They do not overlap with any predicted coding exons, but show clear signs of conserved RNA structures detected by both RNAz and EvoFold with additional support of AlifoldZ. The structure models show a series of consistent/compensatory mutations, and the RNA was detected by the RACE experiments. In the case of example D, further support for the intronic region to be part of a stable ncRNA comes from TARs/Transfrags as well as a short EST mapping nearby and overlapping with two additional RNAz and EvoFold hits. Examples F and G (in Fig. 7F,G) show alternative splicing products of two protein loci detected and confirmed by the GENCODE annotation project. In F, we observe an internal transcription start (further supported by a CpG island), which gives rise to a transcript without clear coding potential but that is highly structured: There are five independent RNAz hits, two of which overlap with EvoFold hits and two with significant AlifoldZ scores (–5.0 and –6.4). A similar situation can be observed in G, where high densities of RNAz hits and overlapping EvoFold hits coincide with noncoding transcripts that arise from an alternatively spliced protein–gene locus.
RNA secondary structures can provide important clues that a given locus is probably transcribed and that this transcript is functional at the RNA level. Here we attempted to comprehensively detect functional structures. Due to the lack of generic sequence signals that would imply RNA function, at present the only way toward this goal (apart from functional studies of individual transcripts) is comparative analysis. As the ENCODE regions are deeply sequenced, they provide an ideal proving ground for such an endeavor.
In contrast to previous genome-wide screens for structured RNA, which were restricted to very well-conserved regions of the genome, here we screened all alignable sequences. Indeed, high sequence conservation is not necessarily needed for function (Bentwich et al. 2005
Using our highest threshold level and considering our estimates of false positives on shuffled alignments, we estimate
Despite the rich comparative sequence data in the ENCODE regions, both RNAz and EvoFold exhibit fairly high false discovery rates of 50%–70% as estimated from randomized input data and correction for dinucleotide frequencies. Also, this high noise level reduces the observed overlap. The overlap for previous screens restricted to phastCons conserved regions, for example, resulted in a twofold higher overlap. Substantial noise levels, however, also plague the experimental approaches. For example, tiling arrays, CAGE, and PET diTags techniques show excellent recovery rates and overlap on annotated coding transcripts, but elsewhere result in large numbers of other signals with moderate overlap and of uncertain relevance. The same is true for protein-coding gene prediction, which yields excellent results on known protein-coding exons but also predict thousands of additional exons incorrectly (Guigo et al. 2006 About 25% of a manual selection of ncRNA candidates were verified by means of RT-PCR, indicating that our computational approach detects a significant number of verifiable transcripts. Small and highly structured known ncRNAs are poorly recovered, indicating that the RT-PCR data most likely underestimate the true extent of transcription. In line with the observation from the ENCODE Pilot Project (The ENCODE Project Consortium 2007), we furthermore expect that most noncoding transcripts have a specific spatiotemporal expression pattern; our screen of six tissues is thus a priori expected to have only limited sensitivity.
One can consider various modes of function for noncoding transcripts like transcriptional interference (Martens et al. 2004 We found evidence for functional RNA structures in all regions of the genome. A fraction of these signals is likely to correspond to small ncRNAs in the classical sense, which are processed from introns or transcribed from intergenic regions with dedicated promoters, as is known for snoRNAs or miRNAs. We also found many signals in UTRs (particularly enriched in 3'-UTRs) of well-known protein-coding genes, suggesting regulatory functions of these signals at the mRNA level.
Our computational data, as well as the results from high-throughput experiments and the evidence from individual experimental results, strongly suggest that the functional spectrum of ncRNAs is much broader than previously expected. For example, we have convincing evidence for functional RNA structures in a few dozen coding exons. These might have regulatory roles for the mRNA, but it is also conceivable that they serve a double role as mRNA and ncRNA. Indeed, there is one example with such a dual role described in the literature, the steroid receptor activator (SRA) (Lanz et al. 2002
Our data in combination with other ENCODE data and aided by visualization methods (Kent et al. 2002
Multiple sequence alignments We used 28-way TBA/MultiZ alignments with human (hg17) as the reference sequence, which were provided by the ENCODE alignment group (Margulies et al. 2007 ). We chose the TBA/MultiZ method alignments mainly because all previous applications of the programs used were done on TBA/MultiZ alignments or other alignments constructed from BLASTZ-based local comparisons. None of the three programs used for our analysis can handle unprocessed genome-wide alignments as presented by TBA/MultiZ. A series of preprocessing and filtering steps was necessary. The analysis pipeline varies in detail to meet the specific requirements of the three programs.
RNAz predictions
AlifoldZ predictions
EvoFold predictions
Randomization of alignments
Comparison with other ENCODE data
Selection of candidates for RT-PCR verification The criteria that were used for selecting candidates include high RNAz/AlifoldZ and/or EvoFold scores, absence of any indication of alignment errors or other alignment artifacts, presence of compensatory mutations, genomic location in either introns of protein-coding transcripts, or unannotated intergenic regions. We routinely generated structure annotated and colorized alignments of all hits visualizing the predicted structure together with the mutational pattern. Inspection of the alignments can help to select more reasonable candidates mainly by weeding out obvious false positives. For example, unusual gap patterns or low complexity runs of single letters indicate an artifactual hit. Currently, the programs themselves cannot efficiently recognize such artifacts, and there is still much room for improvement (e.g., by using an explicit indel model in EvoFold). Negative controls were obtained by randomizing the set of ncRNA target regions using the "Random Intervals" tool of the Galaxy2ENCODE system. From the resulting randomized locations, we chose 38 targets: 19 in intergenic regions (nine overlapping TARs/Transfrags) and 19 in intronic regions (nine overlapping TARs/Transfrags). As positive controls, we randomly chose eight regions in exons of mRNAs of known protein-coding genes and the eight ncRNAs from Table 3.
RT-PCR
5'-RACE/array analysis
Data availability
We thank Lukas Endler for discussion and David Haussler for valuable comments on the manuscript. We acknowledge funding from the Austrian GEN-AU projects "noncoding RNA" and "Bioinformatics Integration Network" (to I.L.H.), the DFG Bioinformatics Initiative BIZ-6/1-2 (to P.F.S.), the Danish Research Council [#272-05-0319], the National Cancer Institute (both to J.S.P.), a Marie Curie Outgoing International Fellowship (to J.O.K.), ENCODE grants from National Human Genome Research Institute (NHGRI)/National Institutes of Health (NIH) (especially to the following ENCODE subgroups: Yale [#U01HG03156], Affymetrix, Inc. [#U01HG03147], and GENCODE [#U01HG03150]), the Swiss National Science Foundation (to S.E.A. and to A.R.), the NCCR Frontiers in Genetics and the European Union (to S.E.A.), the Jérôme Lejeune (to S.E.A. and A.R.), the Childcare (to S.E.A.), and the Novartis (to A.R.) Foundations.
14 Corresponding author.
E-mail wash{at}tbi.univie.ac.at; fax 43-1-4277-52793. [The sequenced fragments of verified ncRNA predictions and TEC were deposited to GenBank under accession nos. EF212232–EF212281 and EF212282–EF212289, respectively.] Article is online at http://www.genome.org/cgi/doi/10.1101/gr.5650707
15 This approach is also similar in spirit to QRNA, a program that detects conserved RNA structures in pairwise alignments by comparing an SCFG-based RNA model to a background model (Rivas and Eddy 2001
16 MIRN483 does not overlap with TARs/Transfrags. It might be specific in fetal liver tissue, which is not among the 11 tissues tested.
Bentwich, I., Avniel, A.A., Karov, Y., Aharonov, R., Gilad, S., Barad, O., Barzilai, A., Einat, P., Einav, U., Meiri, E., et al. 2005. Identification of hundreds of conserved and nonconserved human microRNAs. Nat. Genet. 37: 766–770.[CrossRef][Medline] Bertone, P., Stoc, V., Royce, T.E., Rozowsky, J.S., Urban, A.E., Zhu, X., Rinn, J.L., Tongprasit, W., Samanta, M., Weissman, S., et al. 2004. Global identification of human transcribed sequences with genome tiling arrays. Science 306: 2242–2246. Blanchette, M., Kent, W.J., Riemer, C., Elnitski, L., Smit, A.F., Roskin, K.M., Baertsch, R., Rosenbloom, K., Clawson, H., Green, E.D., et al. 2004. Aligning multiple genomic sequences with the threaded blockset aligner. Genome Res. 14: 708–715. Blankenberg, D., Taylor, J., Schenck, I., He, J., Zhang, Y., Ghent, M., Veeraraghavan, N., Albert, I., Miller, W., Makova, K., et al. 2007. A framework for collaborative analysis of ENCODE data: Making large-scale analyses biologist-friendly. Genome Res. (this issue) doi: 10.1101/gr.5578007. Bompfünewerer, A.F., Flamm, C., Fried, C., Fritzsch, G., Hofacker, I.L., Lehmann, J., Missal, K., Mosig, A., Müller, B., Prohaska, S.J., et al. 2005. Evolutionary patterns of non-coding RNAs. Theory Biosci. 123: 301–369.[CrossRef] Buratti, E. and Baralle, F.E. 2004. Influence of RNA secondary structure on the pre-mRNA splicing process. Mol. Cell. Biol. 24: 10505–10514. Carninci, P., Kasukawa, T., Katayama, S., Gough, J., Frith, M., Maeda, N., Oyama, R., Ravasi, T., Lenhard, B., Wells, C., et al. 2005. The transcriptional landscape of the mammalian genome. Science 309: 1559–1563. Chamary, J.V. and Hurst, L.D. 2005. Evidence for selection on synonymous mutations affecting stability of mRNA secondary structure in mammals. Genome Biol. 6: R75.[CrossRef][Medline] Cheng, J., Kapranov, P., Drenkow, J., Dike, S., Brubaker, S., Patel, S., Long, J., Stern, D., Tammana, H., Helt, G., et al. 2005. Transcriptional maps of 10 human chromosomes at 5-nucleotide resolution. Science 308: 1149–1154. Chooniedass-Kothari, S., Emberley, E., Hamedani, M.K., Troup, S., Wang, X., Czosnek, A., Hube, F., Mutawe, M., Watson, P.H., Leygue, E., et al. 2004. The steroid receptor RNA activator is the Clote, P., Ferre, F., Kranakis, E., and Krizanc, D. 2005. Structural RNA has lower folding energy than random RNA of the same dinucleotide frequency. RNA 11: 578–591. Denoeud, F., Kapranov, P., Ucla, C., Frankish, A., Castelo, R., Drenkow, J., Lagarde, J., Alioto, T., Manzano, C., Chrast, J., et al. 2007. Prominent use of distal 5' transcription start sites and discovery of a large number of additional exons in ENCODE regions. Genome Res. (this issue) doi: 10.1101/gr.5660607. The ENCODE Project Consortium, 2007. identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature (in press). Feng, J., Bi, C., Clark, B.S., Mady, R., Shah, P., and Kohtz, J.D. 2006. The evf-2 noncoding RNA is transcribed from the dlx-5/6 ultraconserved region and functions as a dlx-2 transcriptional coactivator. Genes & Dev. 20: 1470–1484. Gabory, A., Ripoche, M.A., Yoshimizu, T., and Dandolo, L. 2006. The H19 gene: Regulation and function of a noncoding RNA. Cytogenet. Genome Res. 113: 188–193.[CrossRef][Medline] Gardner, P.P. and Giegerich, R. 2004. A comprehensive comparison of comparative RNA structure prediction approaches. BMC Bioinformatics 5: 140.[CrossRef][Medline] Glusman, G., Qin, S., El-Gewely, M.R., Siegel, A.F., Roach, J.C., Hood, L., and Smit, A.F. 2006. A third approach to gene prediction suggests thousands of additional human transcribed regions. PLoS Comput. Biol. 2: e18.[CrossRef][Medline] Griffiths-Jones, S., Moxon, S., Marshall, M., Khanna, A., Eddy, S.R., and Bateman, A. 2005. Rfam: Annotating non-coding RNAs in complete genomes. Nucleic Acids Res. 33: D121–D124. Guigo, R., Flicek, P., Abril, J.F., Reymond, A., Lagarde, J., Denoeud, F., Antonarakis, S., Ashburner, M., Bajic, V.B., Birney, E., et al. 2006. EGASP: The human ENCODE Genome Annotation Assessment Project. Genome Biol. (Suppl 1) 7: S2.1–S2.31. Harrow, J., Denoeud, F., Frankish, A., Reymond, A., Chen, C.K., Chrast, J., Lagarde, J., Gilbert, J.G., Storey, R., Swarbreck, D., et al. 2006. GENCODE: Producing a reference annotation for ENCODE. Genome Biol. (Suppl 1) 7: S4.1–S4.9. Hofacker, I.L., Fekete, M., and Stadler, P.F. 2002. Secondary structure prediction for aligned RNA sequences. J. Mol. Biol. 319: 1059–1066.[CrossRef][Medline] Hubert, N., Walczak, R., Sturchler, C., Myslinski, E., Schuster, C., Westhof, E., Carbon, P., and Krol, A. 1996. RNAs mediating cotranslational insertion of selenocysteine in eukaryotic selenoproteins. Biochimie 78: 590–596.[Medline] Hull Havgaard, J.H., Lyngsø, R., Stormo, G.D., and Gorodkin, J. 2005. Pairwise local structural alignment of RNA sequences with sequence similarity less than 40%. Bioinformatics 21: 1815–1824 |
Top
Abstract
Results
rst functional RNA encoding a protein. FEBS Lett. 566: 43–47.