- Open Access
MicroTar: predicting microRNA targets from RNA duplexes
BMC Bioinformatics volume 7, Article number: S20 (2006)
The accurate prediction of a comprehensive set of messenger RNAs (targets) regulated by animal microRNAs (miRNAs) remains an open problem. In particular, the prediction of targets that do not possess evolutionarily conserved complementarity to their miRNA regulators is not adequately addressed by current tools.
We have developed MicroTar, an animal miRNA target prediction tool based on miRNA-target complementarity and thermodynamic data. The algorithm uses predicted free energies of unbound mRNA and putative mRNA-miRNA heterodimers, implicitly addressing the accessibility of the mRNA 3' untranslated region. MicroTar does not rely on evolutionary conservation to discern functional targets, and is able to predict both conserved and non-conserved targets. MicroTar source code and predictions are accessible at http://tiger.dbs.nus.edu.sg/microtar/, where both serial and parallel versions of the program can be downloaded under an open-source licence.
MicroTar achieves better sensitivity than previously reported predictions when tested on three distinct datasets of experimentally-verified miRNA-target interactions in C. elegans, Drosophila, and mouse.
MicroRNAs (miRNAs) are a class of endogenous, small regulatory RNA averaging 22 nucleotides in length that mediate the post-transcriptional regulation of messenger RNAs. They bind to target messages in a sequence-specific manner, and induce translational repression or endonucleolytic cleavage. The first two miRNAs, lin-4 and let-7 were discovered some seven years apart in the worm C. elegans, in genetic screens for mutants with disrupted developmental timing [1, 2]. There has since been an explosion of interest in the field, and the identification of hundreds of miRNAs in metazoans as disparate as vertebrates, arthropods, nematodes, and viruses  has established miRNAs as pervasive regulators of gene expression. For recent reviews, see [4–6].
Functions have only been experimentally assigned to a small fraction of the few thousand known miRNAs . Of the experimental strategies available to investigate miRNA function, stringent genetic tests that link miRNA loss-of-function mutants to misregulated targets, and point mutations in miRNA binding sites to specific phenotypes are impractical on a genomic scale in any animal species . Tissue-culture assays using reporter gene constructs fused to target sequences are an easier alternative, but their reliance on ectopic miRNA expression harbours the danger of measuring what may be a nonphysiological interaction between two molecules with complementary surfaces . Computational approaches are thus likely to remain an important means of studying miRNA targets for the forseeable future, not least as a means of directing wet-lab experiments. These predictions are no doubt hampered by the fact that animal miRNAs – in contrast to plant miRNAs – tend to be only partially complementary to their target mRNAs. This fact, compounded by the small size of these molecules, precludes the use of standard sequence comparison methods.
Several algorithms have been developed to predict miRNA targets in animal species; these are listed in Table 1. A common strategy in several of these programs is to rank target 3' untranslated region (UTR) complementarity by some combination of duplex free energy and/or pairing requirements at the 5' end (seed region) of the miRNA . For instance, TargetScan  combines requirements for conserved perfect Watson-Crick pairing at positions 2–8 of the miRNA with estimates of the free energy of isolated miRNA-target site interactions, ignoring initiation free energy. While in vitro tests have shown sites containing G:U base-pairs to be functional but impaired , recent in vivo experiments have demonstrated them to be efficiently downregulated . Taken together with the presence of a G:U base-pair in the seed region of a functional let-7 binding site in the lin-41 3'-UTR , these results make a case for the inclusion of seeds with G:U wobbles in target prediction algorithms.
The PicTar [13, 14] algorithm defines seeds as heptamers with Watson-Crick or G:U pairings at positions 1–7 or 2–8 from the miRNA 5' end. It combines seed searches with RNA duplex free energy filters, evolutionary conservation requirements, and a probabilistic scoring mechanism to predict targets that are under combinatorial control by co-expressed miRNAs. However, it makes use of RNAHybrid , an algorithm that approximates RNA duplex free energies by discarding intramolecular hybridizations in order to achieve linear time complexity.
Robins et al.  incorporate mRNA secondary structure computed from 3'-UTRs in their target prediction algorithm, but require perfect Watson-Crick complementarity in the seed site. Furthermore, the use of isolated 3'-UTRs is likely to produce structures very different from the structure of 3'-UTRs in folds that use complete mRNA sequences.
While most of the tools listed in Table 1 are accessible as web services, only miRanda  and RNAHybrid are available as downloadable software that can be modified, extended and run on custom datasets. Most listed algorithms also rely on target conservation across two or more species as a filter. While this is necessary to distinguish functional targets from a vast array of candidates, it results in the unavoidable omission of real targets that are not thus conserved.
Here we present MicroTar, an miRNA target prediction program that does not rely on evolutionary conservation. Through the use of the partial complementarity of miRNAs to their target messages, and the predicted free energy of complete mRNA molecules, we are able to address the problem of the prediction of targets that are not conserved across different genomes. Moreover, harnessing the power of parallel computing obviates the need for introducing approximations that discard intramolecular base pairs in estimates of miRNA-mRNA duplex free energy; we thus implicitly incorporate the accessibility of 3'-UTRs in the algorithm. MicroTar source code – available under an open-source licence – and predictions can be accessed at the MicroTar website .
The MicroTar algorithm is based on the following assumptions:
miRNA target specificity is determined by a heptameric seed sequence (beginning at the first or second position from the 5' end of the miRNA) that is complementary to sites in mRNA 3'-UTRs
targets are functional if miRNA-mRNA duplex formation is energetically favourable
Beginning with a set of fasta-formatted query (miRNA) sequences and target (mRNA) sequences, the MicroTar algorithm predicts the minimum free energy of the each mRNA molecule, searches for seed sites, and performs a constrained fold where each seed match is, in turn, bound in the miRNA-mRNA heterodimer; the output is a list of putative duplexes more stable than free mRNA, along with images of bound and unbound mRNA secondary structure. This result is subsequently subjected to a statistical analysis to determine the significance of each miRNA-mRNA match. Figure 1 presents a schematic overview of this algorithm.
Secondary structure prediction
The secondary structure and minimum free energy of the complete unbound mRNA molecule are predicted using the fold routine from the RNAlib library of the ViennaRNA package . This is an implementation of the Zuker & Stiegler dynamic programming algorithm . We denote the predicted free energy of unbound mRNA as G1.
Loss-of-function mutation studies have demonstrated the core of miRNA sequence specificity to be a heptameric seed sequence , which we define as nucleotides 1–7 or 2–8 at the 5' end of the miRNA. MicroTar searches each mRNA 3'-UTR (or complete mRNA in the absence of annotations) for sites with Watson-Crick or G–U wobble complementarity to this seed sequence; we refer to these hits as seed matches.
For each seed match above, the mRNA is again folded under the constraint that the miRNA seed is bound to its corresponding match. This uses the cofold  routine from the RNAlib library. We denote the free energy of the duplex as G2.
The output is a list of all seed matches, along with predicted energies of the unbound mRNA (G1), putative mRNA-miRNA heterodimers (G2), the estimated energy of duplex formation (g = G2 - G1), and optionally, images of the secondary structure of each mRNA before and after miRNA binding (see e.g., Figure 2).
Seed matches are considered functional targets if the relevant miRNA-mRNA heterodimer is more energetically stable than free mRNA, i.e., g < 0. We then estimate the significance of the prediction using extreme value statistics, much in the fashion of Rehmsmeier et al. . This procedure is outlined below.
Statistical analysis of predicted targets
Negative normalized free energy
The occurrence of favourable hybridizations of short miRNAs with long mRNAs can frequently be attributed to chance: the longer the mRNA, the more likely the incidence. In order to eliminate the effect of sequence length on our measure of free energy [15, 22], we define the negative normalized free energy
where m is the length of the target sequence searched, and n is the length of the miRNA.
Extreme value statistics
Extreme value distributions (EVDs) are limiting distributions that describe the minimum or maximum of independent random variables . If we consider the miRNA-mRNA duplex energy estimation to be essentially an optimization procedure that produces a minimum, the negative normalized free energy described above is a corresponding maximum, and can be described by an EVD having a distribution function of the form
A transformation then converts this distribution function into a straight line:
By scanning for targets of random miRNA sequences in the mRNA sequences in the dataset, we obtain a set of negative normalized free energies, which we expect will follow an EVD. We then transform the distribution function of the empirical EVD into a straight line, as in Equation 3, and estimate the parameters of the EVD by a linear least squares fit to the line y = mx + c, obtaining
a = cb. (5)
We can now compute, for each predicted miRNA-mRNA duplex, a p-value, the probability that the same or a more favourable free energy is observed due to chance:
where a and b are estimated EVD parameters, and g n is the negative normalized free energy from Equation 1 .
MicroTar has been written using the C programming language, and makes use of the RNAlib library from the Vienna RNA package . Great care has been taken to make the system suitable for datasets of varying sizes. Sequences are loaded into memory only as required, allowing the handling of virtually any number of sequences. The parallel version uses functions from v2.0 of the Message Passing Interface (MPI).
MicroTar should compile and run under Linux and most flavours of UNIX. It has been tested under Fedora Core 4 & 5 and CentOS 4.4 Linux distributions, on both 32 and 64 bit platforms.
Results and Discussion
We performed a test of MicroTar on three sets of experimentally verified miRNA targets in C. elegans, Drosophila, and mouse, from v3.0 of TarBase . miRNA sequences were retrieved from miRBase v9.0 ; mRNA sequences from RefSeq entries associated with the corresponding gene entry in the Entrez Gene database. In the absence of 3'-UTR annotations, the entire mRNA sequence was scanned for seed matches by MicroTar. These results are summarized in Figure 3, which shows a density plot of free energies of the most stable predicted miRNA-target duplex for each gene-miRNA pair in the three species.
Furthermore, we compared our predictions to the widely-used PicTar algorithm, which was recently updated and applied to miRNAs in C. elegans. This comparison is shown in Table 2, where we note that MicroTar achieves better sensitivity in all three cases. We emphasize that unverified predicted interactions should be viewed as a guide for further experiments and not as false positives. Detailed lists of targets predicted are available as supplementary data (see Additional File 1 – MicroTar target predictions compared to PicTar), and on the MicroTar website .
Duplex energy estimation
At the core of the MicroTar algorithm lies a novel approach to the estimation of miRNA-mRNA duplex energy. Interactions are viewed in a global context by predicting folds for the entire mRNA, rather than just its 3'-UTR or seed match. By allowing intramolecular hybridizations, we implicitly incorporate the accessibility of the 3'-UTR; seed matches in highly inaccessible UTRs are expected to disrupt UTR secondary structure in putative duplexes. Large disruptions in base pairing cannot be compensated for by bond formation during miRNA-mRNA hybridization. This results in a putative duplex with free energy G2 far greater than that of the unbound mRNA, G1, and the match is rejected.
Significance of predictions
In order to estimate the significance of our predictions, we calculated the p-value for the lowest energy duplex for each miRNA-transcript pair, as derived in Equation 6. The parameters were estimated separately for each species from a distribution computed with random miRNAs. We shuffled miRNAs using the shuffleseq utility from the EMBOSS package , ensuring that there were a sufficient number of random sequences for approximately 4000 seed matches in each species. Figure 4 shows these p-values in a density plot for each miRNA-target pair, as in Figure 3.
MicroTar does not rely on evolutionary conservation to filter predicted targets and is able to address the problem of the prediction of targets that are not conserved across different genomes. Parallel computing makes feasible the use of complex energy prediction algorithms on a large scale, and by using estimates of miRNA-mRNA duplex free energy that allow intramolecular pairings, MicroTar implicitly incorporates the accessibility of 3'-UTRs. In tests on three datasets of experimentally verified miRNA targets in C. elegans, Drosophila and mouse, MicroTar displays greater sensitivity than previously developed target prediction programs.
Availability and Requirements
Project name: MicroTar
Project home page: http://tiger.dbs.nus.edu.sg/microtar/
Operating systems: Linux, UNIX
Programming language: C
Other requirements: GNU autoconf/automake
Licence: New BSD licence
Any restrictions to use by non-academics: None (check ViennaRNA licence, however)
Lee RC, Feinbaum RL, Ambros V: The C. elegans heterochronic gene lin-4 encodes small RNAs with antisense complementarity to lin-14. Cell 1993, 75: 843–854. 10.1016/0092-8674(93)90529-Y
Reinhart BJ, Slack FJ, Basson M, Pasquinelli AE, Bettinger JC, Rougvie AE, Horvitz HR, Ruvkun G: The 21-nucleotide let-7 RNA regulates developmental timing in Caenorhabditis elegans. Nature 2000, 403: 901–906. 10.1038/35002607
Griffiths-Jones S, Grocock RJ, van Dongen S, Bateman A, Enright AJ: miRBase: microRNA sequences, targets and gene nomenclature. Nucleic Acids Res 2006, 34: D140-D144. 10.1093/nar/gkj112
Bartel DP: MicroRNAs: Genomics, Biogenesis, Mechanism, and Function. Cell 2004, 116: 281–297. 10.1016/S0092-8674(04)00045-5
Du T, Zamore PD: microPrimer: the biogenesis and function of microRNA. Development 2005, 132: 4645–4652. 10.1242/dev.02070
Kim VN, Nam JW: Genomics of microRNA. Trends Genet 2006, 22: 165–173. 10.1016/j.tig.2006.01.003
Sethupathy P, Corda B, Hatzigeorgiou AG: TarBase: A comprehensive database of experimentally supported animal microRNA targets. RNA 2006, 12: 192–197. 10.1261/rna.2239606
Lai EC: Predicting and validating microRNA targets. Genome Biol 2004, 5: 115. 10.1186/gb-2004-5-9-115
Didiano D, Hobert O: Perfect seed pairing is not a generally reliable predictor for miRNA-target interactions. Nat Struct Mol Biol 2006, 13: 849–851. 10.1038/nsmb1138
Lewis BP, Shih IH, Jones-Rhoades MW, Bartel DP: Prediction of Mammalian MicroRNA Targets. Cell 2003, 115: 787–798. 10.1016/S0092-8674(03)01018-3
Brennecke J, Stark A, Russell RB, Cohen SM: Principles of MicroRNA-Target Recognition. PLoS Biol 2005, 3: e85. 10.1371/journal.pbio.0030085
Vella MC, Choi EY, Lin SY, Reinert K, Slack FJ: The C. elegans microRNA let-7 binds to imperfect let-7 complementary sites from the lin-41 3'UTR. Genes Dev 2004, 18: 132–137. 10.1101/gad.1165404
Krek A, Grün D, Poy MN, Wolf R, Rosenberg L, Epstein EJ, MacMenamin P, da Piedade I, Gunsalus KC, Stoffel M, Rajewsky N: Combinatorial microRNA target predictions. Nat Genet 2005, 37: 495–500. 10.1038/ng1536
Lall S, Grün D, Krek A, Chen K, Wang YL, Dewey CN, Sood P, Colombo T, Bray N, MacMenamin P, Kao HL, Gunsalus KC, Pachter L, Piano F, Rajewsky N: A genome-wide map of conserved microRNA targets in C. elegans. Curr Biol 2006, 16: 460–471. 10.1016/j.cub.2006.01.050
Rehmsmeier M, Steffen P, Höchsmann M, Giegerich R: Fast and effective prediction of microRNA/target duplexes. RNA 2004, 10: 1507–1517. 10.1261/rna.5248604
Robins H, Li Y, Padgett RW: Incorporating structure to predict microRNA targets. Proc Natl Acad Sci U S A 2005, 102: 4006–4009. 10.1073/pnas.0500775102
John B, Enright AJ, Aravin A, Tuschl T, Sander C, Marks DS: Human MicroRNA Targets. PLoS Biol 2004, 2: e363. 10.1371/journal.pbio.0020363
MicroTar: microRNA target prediction[http://tiger.dbs.nus.edu.sg/microtar/]
Hofacker IL, Fontana W, Stadler PF, Bonhoeffer LS, Tacker M, Schuster P: Fast folding and comparison of RNA secondary structures. Monatsh Chem 1994, 125: 167–188. 10.1007/BF00818163
Zuker M, Stiegler P: Optimal computer folding of large RNA sequences using thermodynamics and auxiliary information. Nucleic Acids Res 1981, 9: 133–148. 10.1093/nar/9.1.133
Bernhart SH, Tafer H, Mückstein U, Flamm C, Stadler PF, Hofacker IL: Partition function and base pairing probabilities of RNA heterodimers. Algorithms Mol Biol 2006, 1: 3. 10.1186/1748-7188-1-3
Karlin S, Altschul SF: Methods for assessing the statistical significance of molecular sequence features by using general scoring schemes. Proc Natl Acad Sci U S A 1990, 87: 2264–2268. 10.1073/pnas.87.6.2264
Gumbel EJ: Statistics of Extremes. New York: Columbia University Press; 1958.
Rice P, Longden I, Bleasby A: EMBOSS: The European Molecular Biology Open Software Suite. Trends Genet 2000, 16: 276–277. 10.1016/S0168-9525(00)02024-2
Rusinov V, Baev V, Minkov IN, Tabler M: MicroInspector: a web tool for detection of miRNA binding sites in an RNA sequence. Nucleic Acids Res 2005, 33: W696-W700. 10.1093/nar/gki364
Kiriakidou M, Nelson PT, Kouranov A, Fitziev P, Bouyioukos C, Mourelatos Z, Hatzigeorgiou A: A combined computational-experimental approach predicts human microRNA targets. Genes Dev 2004, 18: 1165–1178. 10.1101/gad.1184704
Sætrom O, Ola Snøve J, Sætrom P: Weighted sequence motifs as an improved seeding step in microRNA target prediction algorithms. RNA 2005, 11: 995–1003. 10.1261/rna.7290705
Stark A, Brennecke J, Russell RB, Cohen SM: Identification of Drosophila microRNA targets. PLoS Biol 2003, 1: e60. 10.1371/journal.pbio.0000060
This work was supported in part by grant R-154-000-265-112 from the National University of Singapore.
RT acknowledges support from the National University of Singapore Research Scholarship.
This article has been published as part of BMC Bioinformatics Volume 7, Supplement 5, 2006: APBioNet – Fifth International Conference on Bioinformatics (InCoB2006). The full contents of the supplement are available online at http://www.biomedcentral.com/1471-2105/7?issue=S5
MTT and RT planned the project. RT acquired the data and implemented the algorithm. Both authors prepared and approved the final manuscript.
Electronic supplementary material
Additional File 1: . A list of all experimentally verified targets in the three datasets used (C. elegans, Drosophila and mouse), with a comparison of those predicted by MicroTar and those found on the PicTar website. (XLS 34 KB)
About this article
Cite this article
Thadani, R., Tammi, M.T. MicroTar: predicting microRNA targets from RNA duplexes. BMC Bioinformatics 7, S20 (2006). https://doi.org/10.1186/1471-2105-7-S5-S20
- Free Energy
- miRNA Target
- miRNA Target Prediction
- mRNA Secondary Structure
- Seed Match