Bioinformatics Advance Access originally published online on July 26, 2006
Bioinformatics 2006 22(19):2320-2325; doi:10.1093/bioinformatics/btl394
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Prediction of mRNA polyadenylation sites by support vector machine
1 Department of Mathematical Sciences, New Jersey Institute of Technology Newark, NJ 07102, USA
2 Department of Biochemistry and Molecular Biology, New Jersey Medical School University of Medicine and Dentistry of New Jersey, Newark, NJ 07101, USA
*To whom correspondence should be addressed.
 |
ABSTRACT |
|---|
 TOP ABSTRACT 1 INTRODUCTION2 METHODS3 RESULTSREFERENCES |
|---|
mRNA polyadenylation is responsible for the 3' end formation of most mRNAs in eukaryotic cells and is linked to termination of transcription. Prediction of mRNA polyadenylation sites [poly(A) sites] can help identify genes, define gene boundaries, and elucidate regulatory mechanisms. Current methods for poly(A) site prediction achieve moderate sensitivity and specificity. Here, we present a method using support vector machine for poly(A) site prediction. Using 15 cis-regulatory elements that are over-represented in various regions surrounding poly(A) sites, this method achieves higher sensitivity and similar specificity when compared with polyadq, a common tool for poly(A) site prediction. In addition, we found that while the polyadenylation signal AAUAAA and U-rich elements are primary determinants for poly(A) site prediction, other elements contribute to both sensitivity and specificity of the prediction, indicating a combinatorial mechanism involving multiple elements when choosing poly(A) sites in human cells.
Contact: btian{at}umdnj.edu
|
1 INTRODUCTION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTSREFERENCES |
|---|
mRNA polyadenylation is the cellular process that adds poly(A) tails to maturing mRNAs. The process of polyadenylation is composed of two tightly coupled steps (Colgan and Manley, 1997): an endonucleolytic cleavage at a polyadenylation site [poly(A) site] and subsequent polymerization of an adenosine tail at the 3' end of cleaved RNA. Polyadenylation is directly linked to the termination of transcription (Buratowski, 2005; Proudfoot, 2004). Malfunction of polyadenylation has been implicated in several human diseases (Bennett et al., 2001; Gehring et al., 2001).
Signals required for promoting polyadenylation reside near the poly(A) site. The genomic sequence surrounding a poly(A) site is referred to as a poly(A) region. The nucleotide composition of human poly(A) regions is generally U-rich (Legendre and Gautheret, 2003; Tian et al., 2005). A hexamer AAUAAA or a close variant, usually referred to as the polyadenylation signal (PAS), is located 10–35 nt upstream of most human poly(A) sites (Tian et al., 2005). U/GU-rich sequences are located within 
40 nt downstream of the poly(A) sites (Hu et al., 2005; Zarudnaya et al., 2003). In addition, a number of auxiliary upstream elements (USE) or downstream elements (DSE) have been identified in viral and cellular systems [Hu et al. (2005) and references therein]. Yeast and plant genes utilize a distinct set of cis-regulatory elements, or cis elements, for polyadenylation (Graber et al., 1999; Zhao et al., 1999). While AAUAAA is a prominent hexamer located upstream of poly(A) sites in these species, it occurs to a much lesser extent than in mammalian systems. Other A-rich elements seem to be equivalent to AAUAAA. In addition, UAUA and UGUA elements are the efficiency elements (EEs) located 30–70 nt upstream of yeast poly(A) sites (Graber, 2003), which also have been found to be functional elements in human cells (Venkataraman et al., 2005).
Prediction of poly(A) sites has been attempted by several groups during the last several years. An early approach by Salamov and Solovyev (1997) used a linear discriminant function. A number of variables were used, including position weight matrices for the upstream AAUAAA element and downstream U/GU-rich element, distance between AAUAAA and U/GU-rich elements, and hexamer and triplet compositions in both upstream and downstream regions. Tabaska and Zhang (1999) developed polyadq, which employed two quadratic discriminant functions for sequences containing AAUAAA and AUUAAA. The program also uses a position weight matrix for the downstream sequence, a weighted average of hit positions for DSE, and downstream dimer preferences. In addition, weight-matrix-only (Legendre and Gautheret, 2003) and hidden Markov model (HMM) approaches (Graber et al., 2002; Hajarnavis et al., 2004) also have been employed for poly(A) site prediction. Overall, current methods achieve moderate sensitivity and specificity.
Using a hexamer enrichment method called PROBE, we recently identified 15 cis elements in 4 regions surrounding human poly(A) sites (Hu et al., 2005) (Fig. 1). These cis elements were suggested to play enhancing roles in mRNA polyadenylation, as they (1) are over-represented in human poly(A) regions compared with random sequences, and (2) have higher frequency of occurrence for frequently used poly(A) sites than for less frequently used ones. Based on their locations, elements are named as follows: elements in the –100 to –41 nt region are called AUEs (auxiliary USE, four in total); elements in the –40 to –1 nt region are called CUEs (core USE, two in total); elements in the +1 to +40 nt region are called CDEs (core DSE, four in total); and elements in the +41 to +100 nt region are called ADEs (auxiliary USE, five in total). Here, we present a method using support vector machine (SVM) for poly(A) site prediction by these 15 cis elements. This method, which is implemented in a program called polya_svm, achieved higher sensitivity and similar specificity when compared with polyadq. In addition, we found that while PAS and U-rich elements are the most important determinants for prediction, other elements also contribute to the sensitivity and specificity of prediction, indicating a combinatorial mechanism involving multiple elements when utilizing poly(A) sites in human cells.
|
|
2 METHODS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTSREFERENCES |
|---|
2.1 Datasets
We used 29 283 human genomic sequences surrounding the poly(A) sites (–300 to +300 nt) in the polyA_DB database (Zhang et al., 2005), which correspond to 13 942 genes. Poly(A) sites in the polyA_DB database were identified by aligning cDNA/ESTs with genome sequences using a method described in Tian et al. (2005). The training set for linear discriminant analysis (LDA), quadratic discriminant analysis (QDA) and SVM consisted of 4000 sequences, with 2000 positive sequences randomly selected from polyA_DB and 2000 negative sequences generated by a first-order Markov chain model derived from the positive sequences. To test if there is a bias in choosing the training data, we randomly generated training data 10 times and did not observe any difference (data not shown). The chromosome 1 sequence of the human genome (hg17 version) was downloaded from the UCSC genome bioinformatics site (http://genome.ucsc.edu). RefSeq sequences (August 2005 version) were obtained from NCBI (Pruitt and Maglott, 2001). Poly(A) site groupings based on usage and location were carried out as in (Zhang et al., 2005).
2.2 Scoring 15 cis elements
Position-specific scoring matrices (PSSMs) of 15 cis elements (Hu et al., 2005) were used to search training or testing sequences, and the score for each matching sub-sequence of n nt was calculated by 
where mi,j is the score of nucleotide i at position j in the PSSM. If the value is infinite, i.e. no possibility of occurrence, then S was set to –15, as it was the lowest value we observed in our training data. For each element, we used the maximum score in its corresponding region for prediction. For example, for CUE2, we used its maximum score in the –40 to –1 nt region. Scores were scaled by (S – 
)/
S where is the mean of S for all positive and negative sequences in the training data, and S is the standard deviation of S for all positive and negative sequences in the training data.
2.3 Machine learning
Testing of LDA, QDA and SVM was carried out in program R (http://www.r-project.org) with default settings. For prediction, we used the following equations for Sensitivity (SN), Specificity (SP), and Correlation Coefficient (CC):
Sensitivity: SN = TP/(TP + FN), Specificity: SP = TP/(TP + FP) and 
. TP is true positive, TN is true negative, FN is false negative and FP is false positive.
2.4 Polya_svm
For SVM predictions, we used the SVM library LIBSVM (www.csie.ntu.edu.tw/~cjlin/libsvm), and applied the C-support vector classification (C-SVC) method and the radial basis kernel function (RBF), with default settings, i.e. C = 1 and gamma = 1/15. For calculating probabilities, we applied a window-based adjustment method using probability values generated by LIBSVM: The probability of having a poly(A) site at position i is called its E-value and is calculated by Ei = 
where j is a position relative to i, Pr(j) is the probability value obtained from LIBSVM for position j, and j 
{–4, –3, –2, –1, 0, 1, 2, 3, 4, 5}. Thus, the E-value is based on a 10 nt region centered at i, and the higher the probability, the lower the E-value. In addition, we also required that the region (–15/+15) adjacent to a positive site to have Pr(i) > 0.5 at every position of its sequence. The method is implemented in the program polya_svm, which can be downloaded from http://exon.umdnj.edu/polya_svm.
|
3 RESULTS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTSREFERENCES |
|---|
3.1 Fifteen cis elements in human poly(A) regions
We have previously identified 15 candidate cis elements in four regions of human poly(A) sites using a hexamer enrichment approach (Hu et al., 2005) (Fig. 1). In the present study, we wished to address whether these elements could be used to predict poly(A) sites and how each one contributes to the prediction. On one hand, prediction of poly(A) sites with high sensitivity and specificity can help identify genes in the genome, define gene boundaries and elucidate regulatory mechanisms. On the other hand, successful prediction of poly(A) sites using these elements can validate the functions of these elements, and provide insights into how the polyadenylation machinery works in human cells. To this end, we first used PSSMs of 15 elements to search the –100 to +100 nt region of 29 283 human poly(A) sites listed in the polyA_DB database (Zhang et al., 2005). These poly(A) sites were identified using poly(A/T)-tailed cDNA/EST sequences, are located in different regions of a gene, such as introns, internal exons and 3'-most exons, and correspond to 13 942 genes (Tian et al., 2005). As shown in Figure 2, different elements are present in poly(A) regions to various degrees. The PAS element (CUE2), upstream and downstream U-rich elements (AUE2, CUE1, CDE2), UGUA element (AUE4), UGYCU element (CDE1) and G-rich elements (ADE3 and ADE5) tend to be present in poly(A) regions with higher frequency than others. Interestingly, the UGUG element (CDE3), which is the binding site for CstF-64 (Perez Canadillas and Varani, 2003), does not occur extensively.
|
To understand the relationship among cis elements, we applied a hierarchical clustering method to group the 15 cis elements based on their occurrence. Using Pearson Correlation as the metric and average linkage for tree building, the 15 elements can be largely divided into two groups (Fig. 2). One group consisted of CUE1, CUE2, CDE2, AUE2, AUE3 and AUE4, which are all USE except CDE2, and the other consisted of DSE except AUE1. This grouping is robust, as clustering using other parameters, such as Kendall's Tau Correlation and complete linkage, also resulted in similar groupings (data not shown). Thus, USE and DSE in general have different profiles, indicating that they may compensate for each other during mRNA polyadenylation. In biochemical terms, this result suggests that weak upstream signals can be ‘helped’ by strong downstream signals, and vice versa. However, further experiments are needed to confirm this model.
3.2 Prediction of poly(A) site by SVM
Naturally, 15 cis elements can be considered as 15 variables and used in machine-learning tools for poly(A) site prediction. In particular, methods that take into account interactions between the variables are most suitable for predicting poly(A) sites. This is because cis elements are recognized by RNA-binding proteins during mRNA polyadenylation, such as CPSF-160 binding to AAUAAA, CF Im binding to the UGUA element, CstF-64 binding to the U-rich and UG-rich elements, hFip1 binding to the U-rich element, and hnRNP H family proteins binding to the G-rich element, and extensive protein–protein interactions have been reported for proteins in the polyadenylation machinery (Proudfoot, 2004). In light of this, we tested three discriminant analysis methods, namely LDA, QDA and SVM. LDA finds a hyper-plane to separate two or more classes with a linear combination of variables (Zhang, 2000). It assumes that the data distribution for each class is normal and that all classes have the same covariance. QDA uses a quadratic surface to separate classes (Zhang, 2000), and also makes the assumption of a normal distribution, but relaxes the requirement of covariance. SVM employs kernel functions to separate data by a hyperplane that is supported by vectors lying at the boundaries of the classes (Cortes and Vapnik, 1995). Several formulations and kernels are available for SVM (Burges and Smola, 1998). All of these methods have been used on biological sequences for identification of signals, such as splice site (Yeo et al., 2004; Zhang, 2000; Zhang et al., 2003).
To compare these methods, we randomly selected 2000 poly(A) sites from the polyA_DB database, retrieved the –100/+100 nt genomic region surrounding each poly(A) site, and used them as a positive dataset. We then randomized the positive sequences by a first-order Markov Chain (MC) model to obtain 2000 negative sequences, each with 200 nt in length. Using LDA, QDA and SVM functions in program R, we compared their performance for prediction of poly(A) sites with respect to Sensitivity (SN), Specificity (SP) and Correlation Coefficient (CC, see Materials and Methods for calculation of these values). As summarized in Table 1, of these three methods, QDA achieved the best sensitivity and SVM achieved the best specificity. Overall, SVM has the best performance judged by CC. Thus, we have selected SVM as the prediction method for further studies.
|
To examine how each element contributes to the prediction and whether or not we can reduce the number of variables, we conducted a leave-one-out experiment, where we left out one element at a time and calculated its effect on SN, SP and CC in poly(A) site prediction. We reasoned that omission of important elements would significantly lower the performance of prediction, whereas omission of non-essential ones would not make much difference. As shown in Table 2, we found that CUE2 is the most important one, as omission of CUE2 led to a substantial drop of both sensitivity and specificity, which is consistent with the notion that the AAUAAA element is critically important for mRNA polyadenylation. Omissions of CUE1 or CDE2 had similar effects, albeit to a lesser extent, indicating that U-rich elements surrounding the poly(A) sites are important determinants. In addition, omissions of AUE3, ADE3 or ADE5 caused drops in sensitivity, indicating their important roles in poly(A) site selection. For the rest of the nine elements, leaving out any single element caused some decrease in prediction performance, while none of them appeared to be significant based on a t-test (P-value > 0.05, Table 2). However, leaving out all nine elements made both sensitivity and specificity drop significantly. Thus, these nine elements may contribute to poly(A) site recognition coordinately and some elements may be important for only a small subset of poly(A) sites. Taken together, these data indicate that the 15 elements are necessary for poly(A) site prediction, validating the functional importance of these elements for polyadenylation. In addition, the fact that multiple variables are required for poly(A) site prediction suggests a combinatorial mechanism for poly(A) site recognition in human cells.
|
3.3 Polya_svm
Encouraged by our initial results using SVM, we developed a stand-alone program named polya_svm for poly(A) prediction using the 15 cis elements and SVM. The program searches an input sequence and uses LIBSVM (www.csie.ntu.edu.tw/~cjlin/libsvm) to make predictions using support vectors derived from our experiments described above. Since the prediction has to be carried out at each position of a given sequence, a multiple testing issue arises when predicting the likelihood of a sequence containing a poly(A) site. Since polyadenylation cleavage is usually heterogeneous and occurs in a window rather than at a defined position (Tian et al., 2005), we designed a window-based scoring scheme to address the multiple testing issue: We required a region M of m nt to have probability >0.5 at each position in the region and another region N of n nt within M to have high probability values. M is called a positive region, and N is called a high probability region (HPR, Fig. 1). Using different combinations of m and n, we found that m = 30 and n = 10 achieved the best performance when the product of the 10 probabilities for HPR was set to be >0.5. Thus, on average, the probability of being positive for each position is >0.933 in HPR. As shown in Figure 3, using this method, polya_svm can effectively eliminate false positive sites and accurately locate real poly(A) sites.
|
We tested polya_svm using all human poly(A) regions in the polyA_DB (29 283 in total) and compared its performance with polyadq, a commonly used tool for poly(A) site prediction (Tabaska and Zhang, 1999). For polya_svm, if the predicted location (middle of HPR) is within 24 nt from a real poly(A) site, the prediction was considered true positive (TP), and otherwise false negative (FN). For polyadq, since it uses the PAS location for poly(A) site prediction, we considered a prediction to be TP if a PAS is within 48 nt upstream of a real poly(A) site. As shown in Table 3, polya_svm is 33.8% more sensitive than polyadq (52.8% SN versus 39.5% SN). We next divided poly(A) sites into different groups based on two criteria: their usage and location. For poly(A) site usage, we used the number of supporting cDNA/ESTs for a poly(A) site to determine its frequency of usage. Poly(A) sites in genes with only one poly(A) site are called constitutive sites. Poly(A) sites in genes with multiple sites were grouped into strong, weak and medium. A strong site is used >75% of the time based on supporting cDNA/ESTs. If a gene has a strong site, other sites are called weak sites. If a gene does not have a strong site, all sites are called medium sites. For poly(A) site location, we first separated poly(A) sites located in introns and internal exons (called upstream sites) from those in 3'-most exons, and then divided poly(A) sites in the 3'-most exons into three groups depending upon their location (Table 3). The 5'-most site is called the first site, the 3'-most site is called the last site, and sites in between the 5'-most and 3'-most sites are called middle sites. In addition, if a 3'-most exon contains only one poly(A) site, the site is called a single site. As shown in Table 3, polya_svm was over 50% more sensitive than polyadq for detecting medium and weak poly(A) sites, and
19.5 and 7.2% more sensitive than polyadq for strong and constitutive poly(A) sites, respectively. As for poly(A) sites located in different regions of a gene, polya_svm also made more sensitive predictions than polyadq in all categories, particularly for poly(A) sites located upstream of the 3'-most poly(A) sites (62.2 and 86.1% more sensitive for the first and middle poly(A) sites in 3'-most exons). A more detailed analysis revealed that the high sensitivity of polya_svm is mainly ascribed to its capability to predict poly(A) sites without AAUAAA or AUUAAA, and sites with weak downstream signals (data not shown). Taken together, these data demonstrate that polya_svm is highly sensitive for poly(A) site prediction.
|
We next examined polya_svm's performance for sequences without poly(A) sites, or negative sequences. A true negative sequence is difficult to obtain, as there is no extensive experimental evidence for negative sequences. Thus, we tested several types of sequences that were presumed to have very few poly(A) sites, including randomized poly(A) regions (–300/+300 nt), randomized genome sequences, mRNA coding sequences (CDS), and 5'-untranslated regions (5'-UTRs). As shown in Table 4, comparable false positives (FP) were predicted by polya_svm and polyadq for randomized sequences, but polya_svm predicts significantly more sites than polyadq in CDS and 5'-UTR sequences (>2-fold). Interestingly, the difference was not significant when randomized CDS and 5'-UTR sequences were used (Table 4), suggesting that some of the false positives in CDS and 5'-UTRs predicted by polya_svm may actually be true positives. Accordingly, it is tempting to speculate that there may, in fact, exist a large number of poly(A) sites in CDS and 5'-UTRs. This would be consistent with previous findings of poly(A) sites in internal exons (Tian et al., 2005; Yan and Marr, 2005). However, this hypothesis has yet to be tested by wet lab experiments. On the other hand, a highly sensitive method would aid in the identification of these sites, which are difficult to detect by cDNA/EST-based approaches, as many of these poly(A) sites would result in aberrant transcripts that may be rapidly degraded by cellular surveillance mechanisms, such as those without an in-frame stop codon (Frischmeyer et al., 2002).
|
In summary, highly sensitive prediction of poly(A) sites was achieved by SVM using 15 cis elements surrounding the poly(A) sites. On the other hand,
47% of positive poly(A) sequences in the polyA_DB database were still predicted to be negative. No obvious difference can be discerned between false negative and true positive sequences with respect to the usage of 15 cis elements (Fig. 2), indicating other unidentified features may account for polyadenylation activity for those false negative sequences, such as RNA structure and genome location. These are to be explored in the future to further improve predictions. |
Acknowledgments |
|---|
The authors thank Michael Zhang for inspirational discussions of the results, and Carol Lutz, Melissa Rogers and members of the B.T. lab for critical reading of the manuscript. Y.C. was supported by an NJIT Presidential Strategic Initiative Scholar Award. B.T. was supported by The Foundation of the University of Medicine and Dentistry of New Jersey.
Conflict of Interest: None declared.
|
FOOTNOTES |
|---|
Associate Editor: Chris Stoeckert
Received on June 12, 2006; accepted on June 12, 2006
|
REFERENCES |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 RESULTSREFERENCES |
|---|
Bennett, C.L., et al. (2001) A rare polyadenylation signal mutation of the FOXP3 gene (AAUAAA
AAUGAA) leads to the IPEX syndrome. Immunogenetics, 53, 435–439[CrossRef][ISI][Medline].
Buratowski, S. (2005) Connections between mRNA 3' end processing and transcription termination. Curr. Opin. Cell Biol, . 17, 257–261[CrossRef][ISI][Medline].
Burges, C.J.C. and Smola, A.J. Advances in Kernel Methods-Support Vector Learning, (1998) , Cambridge, MA MIT Press.
Colgan, D.F. and Manley, J.L. (1997) Mechanism and regulation of mRNA polyadenylation. Genes. Dev, . 11, 2755–2766
Cortes, C. and Vapnik, V. (1995) Support-vector networks. Machine Learning, 20, 273–297.
Frischmeyer, P.A., et al. (2002) An mRNA surveillance mechanism that eliminates transcripts lacking termination codons. Science, 295, 2258–2261
Gehring, N.H., et al. (2001) Increased efficiency of mRNA 3' end formation: a new genetic mechanism contributing to hereditary thrombophilia. Nat. Genet, 28, 389–392[CrossRef][ISI][Medline].
Graber, J.H. (2003) Variations in yeast 3'-processing cis-elements correlate with transcript stability. Trends. Genet, 19, 473–476[CrossRef][ISI][Medline].
Graber, J.H., et al. (1999) In silico detection of control signals: mRNA 3'-end-processing sequences in diverse species. Proc. Natl Acad. Sci. USA, 96, 14055–14060
Graber, J.H., et al. (2002) Probabilistic prediction of Saccharomyces cerevisiae mRNA 3'-processing sites. Nucleic Acids Res, . 30, 1851–1858
Hajarnavis, A., et al. (2004) A probabilistic model of 3' end formation in Caenorhabditis elegans. Nucleic Acids Res, . 32, 3392–3399
Hu, J., et al. (2005) Bioinformatic identification of candidate cis-regulatory elements involved in human mRNA polyadenylation. RNA, 11, 1485–1493
Legendre, M. and Gautheret, D. (2003) Sequence determinants in human polyadenylation site selection. BMC Genomics, 4, 7[Medline].
Perez Canadillas, J.M. and Varani, G. (2003) Recognition of GU-rich polyadenylation regulatory elements by human CstF-64 protein. EMBO. J, 22, 2821–2830[CrossRef][ISI][Medline].
Proudfoot, N. (2004) New perspectives on connecting messenger RNA 3' end formation to transcription. Curr. Opin. Cell Biol, . 16, 272–278[CrossRef][ISI][Medline].
Pruitt, K.D. and Maglott, D.R. (2001) RefSeq and LocusLink: NCBI gene-centered resources. Nucleic Acids Res, . 29, 137–140
Salamov, A.A. and Solovyev, V.V. (1997) Recognition of 3'-processing sites of human mRNA precursors. Comput. Appl. Biosci, . 13, 23–28
Tabaska, J.E. and Zhang, M.Q. (1999) Detection of polyadenylation signals in human DNA sequences. Gene, 231, 77–86[CrossRef][ISI][Medline].
Tian, B., et al. (2005) A large-scale analysis of mRNA polyadenylation of human and mouse genes. Nucleic Acids Res, . 33, 201–212
Venkataraman, K., et al. (2005) Analysis of a noncanonical poly(A) site reveals a tripartite mechanism for vertebrate poly(A) site recognition. Genes. Dev, . 19, 1315–1327
Yan, J. and Marr, T.G. (2005) Computational analysis of 3'-ends of ESTs shows four classes of alternative polyadenylation in human, mouse, and rat. Genome. Res, . 15, 369–375
Yeo, G., et al. (2004) Variation in alternative splicing across human tissues. Genome Biol, . 5, R74[CrossRef][Medline].
Zarudnaya, M.I., et al. (2003) Downstream elements of mammalian pre-mRNA polyadenylation signals: primary, secondary and higher-order structures. Nucleic Acids Res, . 31, 1375–1386
Zhang, H., et al. (2005a) PolyA_DB: a database for mammalian mRNA polyadenylation. Nucleic Acids Res, . 33, D116–D120
Zhang, H., et al. (2005b) Biased alternative polyadenylation in human tissues. Genome Biol, . 6, R100[CrossRef][Medline].
Zhang, M.Q. (2000) Discriminant analysis and its application in DNA sequence motif recognition. Brief Bioinform, . 1, 331–342
Zhang, X.H., et al. (2003) Sequence information for the splicing of human pre-mRNA identified by support vector machine classification. Genome. Res, . 13, 2637–2650
Zhao, J., et al. (1999) Formation of mRNA 3' ends in eukaryotes: mechanism, regulation, and interrelationships with other steps in mRNA synthesis. Microbiol. Mol. Biol. Rev, 63, 405–445
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  Z. Lian, A. Karpikov, J. Lian, M. C. Mahajan, S. Hartman, M. Gerstein, M. Snyder, and S. M. Weissman A genomic analysis of RNA polymerase II modification and chromatin architecture related to 3' end RNA polyadenylation Genome Res., August 1, 2008; 18(8): 1224 - 1237. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Shen, G. Ji, B. J. Haas, X. Wu, J. Zheng, G. J. Reese, and Q. Q. Li Genome level analysis of rice mRNA 3'-end processing signals and alternative polyadenylation Nucleic Acids Res., May 1, 2008; 36(9): 3150 - 3161. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Bianchetti, Y. Wu, E. Guerin, F. Plewniak, and O. Poch SAGETTARIUS: a program to reduce the number of tags mapped to multiple transcripts and to plan SAGE sequencing stages Nucleic Acids Res., September 25, 2007; 35(18): e122 - e122. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Tian, Z. Pan, and J. Y. Lee Widespread mRNA polyadenylation events in introns indicate dynamic interplay between polyadenylation and splicing Genome Res., February 1, 2007; 17(2): 156 - 165. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Y. Lee, I. Yeh, J. Y. Park, and B. Tian PolyA_DB 2: mRNA polyadenylation sites in vertebrate genes Nucleic Acids Res., January 12, 2007; 35(suppl_1): D165 - D168. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||