RNA-RNA interaction prediction using genetic algorithm
- Soheila Montaseri^{1},
- Fatemeh Zare-Mirakabad^{2, 3}Email author and
- Nasrollah Moghadam-Charkari^{4}
https://doi.org/10.1186/1748-7188-9-17
© Montaseri et al.; licensee BioMed Central Ltd. 2014
Received: 19 January 2013
Accepted: 18 June 2014
Published: 29 June 2014
Abstract
Background
RNA-RNA interaction plays an important role in the regulation of gene expression and cell development. In this process, an RNA molecule prohibits the translation of another RNA molecule by establishing stable interactions with it. In the RNA-RNA interaction prediction problem, two RNA sequences are given as inputs and the goal is to find the optimal secondary structure of two RNAs and between them. Some different algorithms have been proposed to predict RNA-RNA interaction structure. However, most of them suffer from high computational time.
Results
In this paper, we introduce a novel genetic algorithm called GRNAs to predict the RNA-RNA interaction. The proposed algorithm is performed on some standard datasets with appropriate accuracy and lower time complexity in comparison to the other state-of-the-art algorithms. In the proposed algorithm, each individual is a secondary structure of two interacting RNAs. The minimum free energy is considered as a fitness function for each individual. In each generation, the algorithm is converged to find the optimal secondary structure (minimum free energy structure) of two interacting RNAs by using crossover and mutation operations.
Conclusions
This algorithm is properly employed for joint secondary structure prediction. The results achieved on a set of known interacting RNA pairs are compared with the other related algorithms and the effectiveness and validity of the proposed algorithm have been demonstrated. It has been shown that time complexity of the algorithm in each iteration is as efficient as the other approaches.
Keywords
RNA secondary structure Minimum free energy Fitness functionBackground
Major successes have been achieved in the treatment of some cancers, including colon, breast and pancreatic by suppressing the gene expression involved in the development of these diseases using RNA-RNA interaction. The interaction between two RNAs is known as the newest and the most efficient method for gene silencing. It has been shown that the small interfering RNAs (siRNAs) can be used for silencing their target mRNAs [1]. Furthermore, small RNAs (sRNAs) play an important role in the regulation of gene expression. They usually bind to their target mRNAs to prevent their translation [2].
In RNA-RNA Interaction Prediction (RRIP) problem, two RNA sequences are given as inputs and the goal is to find the minimum free energy Secondary Structure of Interacting RNAs (SSIR). To tackle this problem, some algorithms have been proposed by research groups. Andronescu et al. [3] proposed a method based on dynamic programming in which two RNA sequences are concatenated as a single sequence and its secondary structure is calculated [3]. Another approach calculates the partition function of a secondary structure complex of multiple interacting RNAs [4]. This method rigorously extends those models of secondary structure to the multi-stranded case. The tools such as RNAhybrid [5], UNAFold [6] and RNAduplex from ViennaRNA package[7] reduce computational time complexity by ignoring all the internal base pairings in both RNAs. RNAup [8, 9] extends the standard partition function approach to RNA secondary structures and employs the single (unpaired) regions on each RNA to find the interaction between them. RNAplex [10, 11] finds the possible hybridization sites for RNA in the large RNA databases based on a slight simplification of the energy model. In this model, the loop energy is assumed to be a function of the loop size.
Recently, a novel algorithm based on the multiple context free grammars was introduced in [12]. Accordingly, two real values called transition and emission probabilities are specified for each rule of the grammar. Then, a derivation tree is constructed for the grammar based on the rules with high probability.
In heuristic based approaches, inRNAs [13] firstly predicts the loop regions in the native structure of each sequence, and then finds the optimal non-conflicting interaction between two RNAs. IntaRNA [14] combines the accessibility of target sites as well as the existence of a user-definable seed to find RNA-RNA interaction. Minimizing the joint free energy between two RNA molecules under a number of energy models with growing complexity was introduced in [2]. Another interesting heuristic approach for this problem was presented in [15]. This algorithm employs some dot matrices representation of all possible base pairs for finding the secondary structure of each RNA and between the two RNAs.
An approximation algorithm was presented in [1], where an RNA-RNA interaction graph is created in which every edge represents a possible bond in or between two RNAs. A set of edges is found to maximize the number of bonds. A statistical sampling algorithm was introduced in [16] based on some modifications to the grammars. It calculates the interaction probabilities for any given single region on RNA. RactIP [17] predicts RNA-RNA interaction using integer programming. Accordingly, it uses the approximate information of the internal and external base pairing probabilities of joint structures as an objective function of integer programming. PETcofold [18] employs covariance information in the internal and external base pairs to predict SSIR of two multiple alignments of RNA sequences. InteRNA [19] reduces the time and space complexity of RRIP problem described by Alkan et al. [2] using dynamic programming sparsification.
One of the pitfalls of the most existing algorithms is their high computational time to predict RNA-RNA interaction, while a number of them have not been performed on some RNA pairs to predict binding sites between two single regions of RNAs. Alkan et al. [2] proved that RNA-RNA interaction prediction is an NP-complete problem.
In this paper, we propose a new genetic algorithm called GRNAs as an appropriate solution for the RRIP problem. This algorithm can be performed on some standard RNA pairs with high accuracy. In this method, at first, all possible stems in each RNA as well as all possible hybrid regions between two RNAs are extracted from a dot matrix. The initial population consists of some individuals, where each of them is an SSIR obtained from some randomly extracted stems and hybrid regions of the dot matrix. The minimum free energy is computed for each individual as a fitness value. For each generation, some individuals are selected to mate based on their fitness values and form a new population. Then, mutation operation is done on a few individuals. The population generation terminates when the free energy of an individual is minimum enough. Finally, one of the best individuals is selected as an optimal SSIR. The algorithm is conducted on some real datasets and compared with some other algorithms to investigate efficiency and validity of the proposed method. The time and space complexity of the proposed method in each iteration is 0(l^{2} + |P|), where l and |P| indicate the sum of the length of two RNAs and the length of an individual, respectively. The results show that the accuracy of the algorithm is as efficient as the other related methods.
The rest of this paper is organized as follows. In Section 2, some definitions and notations are described. In Section 3, a genetic algorithm called GRNAs is presented to predict RNA-RNA interaction. The results and conclusion are discussed in Sections 4 and 5, respectively.
Definitions and notations
An RNA molecule is composed of a long, usually single-stranded chain of nucleotide units; adenine (A), cytosine (C), guanine (G) and uracil (U). Thus, R = r_{1}r_{2} … r_{ n } in 5 ' - 3 ' direction is an RNA sequence, where |R| = n and r_{ i } ∈ {A, C, G, U} (1 ≤ i ≤ n). The RNA structure is formed by the creation of hydrogen bonds between Watson-Crick complementary bases (A - U and C - G) and a Wobble base pair (G - U).
In an RNA secondary structure, each base interacts with at most one other base, and no base pairs cross each other. Two bases r_{ i } and r_{ j } (1 ≤ i < j ≤ n) of the base pair (r_{ i }, r_{ j }) are represented by ' (' and ') ', respectively and each unpaired base is declared by '. '. A stem consists of subsequent base pairs and a loop is one sequence of consecutive unpaired bases.
A secondary structure of two interacting RNAs, R_{1} and R_{2}, contains the set of stems in each RNA and the hybrid regions between two RNAs as well as loops. Each hybrid region consists of subsequent hybrid base pairs between two RNAs. Two bases r_{ i } ∈ R_{1} and r_{ j } ∈ R_{2} of the hybrid base pair (r_{ i }, r_{ j }) are represented by ' [' and '] ', respectively.
In this example, each RNA has one stem. In the left hand RNA, one stem is found by the production of bonding between CGG and the reverse CCG (GCC). There are two hybrid regions between the sequences R_{1} and R_{2}. The first one is produced by binding between UUU and the reverse AAA (AAA). The second one is generated by binding between GGU and the reverse ACC (CCA).
A new genetic algorithm for RNA-RNA interaction prediction
Genetic algorithm is an optimization method based on evolutionary biology that is widely used to solve search and optimization problems [20–22]. In this section, a new genetic algorithm, GRNAs, is presented to predict RNA-RNA interaction. In the following, initial population, fitness function, crossover and mutation operations are introduced.
Initial population
where N is an arbitrary character to distinguish between two sequences and l = m + n + 1.
where d_{1} ∝ d_{2} represents all base pairs in d_{1} overlapping with d_{2}.
Notice that a set of the produced individuals creates an initial population.
Fitness function
where for C ∈ {S, H}, MFE(C) denotes the minimum free energy of structure C. We apply RNAeval.exe [7] to compute minimum free energies of secondary structures and binding sites separately.
Crossover
Crossover operation is performed between the individuals with the rate of 0.9. The good and mediocre individuals are transferred to the next population. The remaining individuals are consecutively selected for crossover operation.
In the following, two new individuals P ' _{1} and P ' _{2} are generated from I ' _{1} and I ' _{2} similar to the described method in the initial population (refers to the Equation 1), respectively.
Mutation
An item x_{ j }, where x_{ j } = 1 is randomly selected and replaced to 0. Then, another x_{ k } (x_{ k }= 0) is replaced to 1. The new individual, P, is obtained from the changed I based on the proposed method in initial population (refers to the Equation 1). Finally, if C_{ k } ∉ P (all the base pairs of d_{ k } have overlapping with the existence sub-diagonals in P), the other x_{ i } (x_{ i } = 0) is selected to replace with 1. This process continues until C_{ k } ∈ P or the defined number of generations is reached. When mutation is performed on a number of individuals, they will be increasingly sorted based on their fitness values.
Termination of the GRNAs algorithm
The GRNAs algorithm terminates when the best individual in definite generations will not be changed or the defined number of generations be reached. After the termination of the algorithm, one of the best individuals is selected as the best folding of two RNAs and the best interaction between the two RNAs.
Time and space complexity
We have obtained the time and space complexity of GRNAs in each iteration. Making the dot matrix needs the complexity of O(l^{2}) where l exposes sum of the length of the two RNAs. Let h and |P| be the number of individuals and the length of an individual P. The time complexity of creating the initial population is O(h. |P| ^{2}). We set h = 40 and |P| = max{|R ' |, |R " |}, so h can be ignored. Sorting individuals based on their fitness values requires O(|P|. h. log h). Crossover and mutation operations take O(|P| ^{2}) and O(|P|), respectively. Thus, the time complexity in each iteration in the proposed algorithm is O(h. |P|(|P| + log h)). The maximum number of iteration is at most I = 20. Therefore, the time complexity of the algorithm is O(I. h. |P|(|P| + log h) + l^{2}) that is simplified with O(l^{2} + |P|).
On the other hand, for storing the h individuals of length |P| we need O(h. |P|) space complexity. Furthermore, the population in the algorithm uses both dot matrix and an array of sub-diagonals. Hence, the storage complexity of these two types is O(l^{2}), where l denotes sum of the length of the two RNAs. Thus the total space complexity of GRNAs is O(h|P| + l^{2}) which is simplified with O(l^{2} + |P|).
Results and discussion
The GRNAs has been performed on a machine with two-Core Intel(R) Duo processor T6670 2.20 GHz and 4 GB RAM to predict the interaction structure between two RNAs. The proposed genetic algorithm is performed on two well-known datasets of RNA-RNA interactions. The first set contains: R1inv-R2inv, Tar-Tar*, DIS-DIS, CopA-CopT and IncRNA_{54}-RepZ in the Escherichia coli bacteria [12]. The joint secondary structures of this dataset include kissing hairpins. We evaluate the performance of joint secondary structure prediction of this dataset.
Also, this algorithm is carried on the second set of datasets with their binding sites including some RNA pairs called: DsrA-Rpos, GcvB-argT, GcvB-dppA, GcvB-gltI, GcvB-livK, GcvB-livJ, GcvB-oppA, GcvB-STM4351, IstR-tisAB, MicA-ompA, MicA-lamB, MicC-ompC, MicF-ompF, OxyS-fhlA, RyhB-sdhD, RyhB-sodB, SgrS-ptsG and Spot42-galK [14]. This dataset is used to appraise the performance of RNA-RNA interaction prediction in binding sites.
The results of joint secondary structure prediction of GRNAs in MCC in comparison to the PETcofold and other joint structure prediction methods such as RNAcofold, inteRNA, Pairfold and RactIP
RNA-RNA pairs | GRNAs | PETcofold | RNAcofold | inteRNA | Pairfold | RactIP |
---|---|---|---|---|---|---|
MicA-ompA | 67 | 87 | 80 | 49 | 86 | 57 |
OxyS-fhlA | 60 | 80 | 61 | 64 | 61 | 48 |
RyhB-uof-fur | 56 | 13 | 21 | 12 | 21 | 19 |
RyhB-sodB | 74 | 67 | 65 | 70 | 65 | 65 |
Average | 64 | 62 | 57 | 49 | 58 | 47 |
Sensitivity | PPV | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
RNA-RNA pairs | GRNAs | inRNAs | IntaRNA | RNAup | RactIP | GRNAs | inRNAs | IntaRNA | RNAup | RactIP |
Tar-Tar* | 100 | 100 | 100 | 100 | 81.5 | 100 | 83.3 | 83.3 | 83.3 | 57.9 |
R1inv-R2inv | 100 | 100 | 100 | 100 | 100 | 100 | 77.8 | 100 | 77.8 | 100 |
DIS-DIS | 100 | 100 | 100 | 100 | 75 | 100 | 100 | 100 | 100 | 78.3 |
CopA-CopT | 89.47 | 88.9 | 100 | 55.6 | 100 | 80.95 | 82.8 | 39.1 | 65.2 | 100 |
IncRNA_{54}-RepZ | 71.44 | 100 | 73.8 | 75 | 100 | 100 | 88.9 | 85 | 85.7 | 83.3 |
DsrA-Rpos | 73.08 | 80.8 | 80.8 | 80.8 | 65.4 | 100 | 77.8 | 77.8 | 77.8 | 73.9 |
GcvB-argT | 87.5 | 95 | 95 | 90 | 95 | 100 | 86.4 | 95 | 94.7 | 100 |
GcvB-dppA | 94.12 | 100 | 100 | 100 | 94.1 | 100 | 85 | 58.6 | 45.9 | 59.3 |
GcvB-gltI | 91.67 | 75 | 0 | 0 | 100 | 95.65 | 50 | 0 | 0 | 100 |
GcvB-livK | 70.83 | 54 | 54.2 | 54.2 | 95.5 | 89.47 | 57 | 56.5 | 56.5 | 95.5 |
GcvB-livJ | 95.45 | 63.4 | 95.5 | 95.5 | 95.8 | 95.45 | 82.4 | 95.5 | 95.5 | 95.8 |
GcvB-oppA | 90.91 | 100 | 100 | 100 | 100 | 100 | 73.3 | 95.7 | 95.7 | 100 |
GcvB-STM4351 | 72 | 76 | 76 | 88 | 88 | 100 | 100 | 90.5 | 95.7 | 100 |
IstR-tisAB | 69.44 | 72.2 | 87.9 | 66.7 | 77.8 | 100 | 100 | 96 | 100 | 100 |
MicA-ompA | 93.75 | 100 | 100 | 100 | 87.5 | 100 | 100 | 100 | 100 | 87.5 |
MicA-lamB | 82.61 | 100 | 100 | 82.6 | 56.5 | 90.48 | 100 | 82.1 | 70.4 | 86.7 |
MicC-ompC | 90.91 | 100 | 100 | 72.7 | 72.7 | 100 | 100 | 53.7 | 41 | 88.9 |
MicF-ompF | 100 | 96 | 96 | 80 | 83.3 | 100 | 96 | 96 | 95.2 | 76.9 |
OxyS-fhlA | 80 | 81.3 | 50 | 37.5 | 56.3 | 86.96 | 100 | 100 | 100 | 81.8 |
RyhB-sdhD | 58.82 | 61.8 | 58.8 | 79.4 | 82.4 | 95.24 | 95.5 | 100 | 79.4 | 82.4 |
RyhB-sodB | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 81.8 | 90 | 39.1 |
SgrS-ptsG | 60.87 | 56.6 | 73.9 | 73.9 | 83.9 | 87.5 | 76.5 | 100 | 100 | 100 |
Spot42-galK | 61.36 | 43.2 | 40.9 | 52.3 | 68.2 | 87.1 | 76 | 64.3 | 52.3 | 69.8 |
Average | 84.1 | 84.5 | 81.9 | 77.6 | 84.7 | 96.03 | 86.5 | 80.5 | 78.4 | 85.1 |
RNA-RNA pairs | GRNAs | inRNAs | IntaRNA | RNAup | RactIP |
---|---|---|---|---|---|
Tar-Tar* | 100 | 90.9 | 90.9 | 90.9 | 67.7 |
R1inv-R2inv | 100 | 87.5 | 100 | 87.5 | 100 |
DIS-DIS | 100 | 100 | 100 | 100 | 76.6 |
CopA-CopT | 85 | 85.7 | 56.2 | 60 | 100 |
IncRNA_{54}-RepZ | 82.92 | 94.1 | 79 | 80 | 90.9 |
DsrA-Rpos | 84.44 | 79.3 | 79.3 | 79.3 | 69.4 |
GcvB-argT | 93.33 | 90.5 | 95 | 92.3 | 97.4 |
GcvB-dppA | 96.97 | 91.9 | 73.9 | 62.9 | 72.2 |
GcvB-gltI | 93.62 | 60 | 0 | 0 | 100 |
GcvB-livK | 76.07 | 55.5 | 55.3 | 55.3 | 95.5 |
GcvB-livJ | 95.45 | 71.7 | 95.5 | 95.5 | 95.8 |
GcvB-oppA | 95.24 | 84.6 | 97.8 | 97.8 | 100 |
GcvB-STM4351 | 83.72 | 86.4 | 82.6 | 91.7 | 93.6 |
IstR-tisAB | 81.96 | 83.9 | 91.8 | 80 | 78.5 |
MicA-ompA | 96.77 | 100 | 100 | 100 | 87.5 |
MicA-lamB | 86.36 | 100 | 90.2 | 76 | 68.4 |
MicC-ompC | 95.24 | 100 | 69.9 | 52.4 | 80 |
MicF-ompF | 100 | 96 | 96 | 86.9 | 80 |
OxyS-fhlA | 83.33 | 89.7 | 66.7 | 54.5 | 66.7 |
RyhB-sdhD | 72.73 | 75 | 74.1 | 79.4 | 82.4 |
RyhB-sodB | 100 | 100 | 90 | 94.7 | 56.3 |
SgrS-ptsG | 71.79 | 65.1 | 85 | 85 | 85 |
Spot42-galK | 72 | 55.1 | 50 | 52.3 | 69 |
Average | 89 | 84.5 | 79.1 | 76.3 | 83.6 |
Our genetic algorithm randomly selects the sub-diagonals to make individuals. Therefore different individuals with variety sub-diagonals are constructed. Due to the nature of proposed genetic approach, some of the RNA-RNA interactions can be predicted more accurate than the other algorithms. For example the accuracy rate of Tar-Tar* is obtained 100%, while maximum accuracy of the other approaches is 90.9%.
Comparison of time and space complexity of some algorithms
Algorithm | Time complexity | Space complexity |
---|---|---|
GRNAs | O(l^{2} + |P|) | O(l^{2} + |P|) |
TIRNA | O(k^{2} log k^{2}) | O(k^{2}) |
SPM | O(n^{3}m^{3}) | O(n^{2}m^{2}) |
LM | O(n^{3}m^{3}) | O(n^{2}m^{2}) |
inRNAs | O(k^{4}w) | O(k^{2}) |
RNAup | O(n^{3}m) | O(n^{2}) |
EBM | O(n^{3}m^{3}) | O(n^{2}m^{2}) |
App | O(n^{3}m^{3}) | O(n^{2}m^{2}) |
Pairfold | O(k^{3}) | O(k^{2}) |
IntaRNA | O(nm + nl^{3}) | O(nm) |
ripalign | O(N^{6}) | O(N^{4}) |
PETcofold | O(MIl^{3}) | |
RactIP | O(n^{5}) |
Conclusion
In this paper, a new genetic algorithm was introduced for solving RNA-RNA interaction prediction problem. In this algorithm, all possible stems in each RNA and hybrid regions between two RNAs are extracted from a dot matrix showing all possible base pairs. Initial population is formed based on some stems and hybrid regions of the dot matrix. Minimum free energy is considered as a fitness function. Crossover operation is done between some consecutive individuals in the population. Mutation is taken on a few randomly selected individuals. Population generation continues until the minimum free energy of the best individual becomes minimal enough. Finally, one of the best individuals is selected to form RNA-RNA interaction structure. The proposed algorithm was tested on several RNA-RNA interaction datasets. The experimental results indicate a high accuracy of GRNAs. Furthermore, time and space complexity of GRNAs is as efficient as the other related studies.
Availability
The program of GRNAs is available at http://mostafa.ut.ac.ir/grnas.
Declarations
Acknowledgment
This research was in part supported by a grant from IPM. Also, we would like to thank Professor Mohammad Ganjtabesh for improving the proposed algorithm of the paper.
Authors’ Affiliations
References
- Mneimneh S: On the approximation of optimal structures for RNA-RNA interaction. Trans Comput Biol Bioinform. 2009, 6: 682-688.View ArticleGoogle Scholar
- Alkan C, Karakoc E, Nadeau JH, Sahinalp C, Zhang K: RNA-RNA interaction prediction and antisense RNA target search. J Comput Biol. 2006, 13: 267-282.View ArticlePubMedGoogle Scholar
- Andronescu M, Zhang ZC, Condon A: Secondary structure prediction of interacting RNA molecules. J Mol Biol. 2005, 345: 987-1001.View ArticlePubMedGoogle Scholar
- Dirks R, Bios J, Schaeffer JM, Winfree E, Pierce N: Thermodynamic Analysis of Interacting Nucleic Acid Strands. Soc Ind Appl Math. 2007, 49: 65-88.Google Scholar
- Rehmsmeier M, Steffen P, Hochsmann M, Giegerich R: Fast and effective prediction of microRNA/target duplexes. RNA. 2004, 10: 1507-1517.View ArticlePubMedPubMed CentralGoogle Scholar
- Markham NR, Zuker M: UNAFold: Software for Nucleic Acid Folding and Hybridization. Methods Mol Biol. 2008, 453: 3-31.View ArticlePubMedGoogle Scholar
- Bernhart S, Tafer H, Mückstein U, Flamm C, Peter F, Stadler P, Hofacker I: Partition Function and Base Pairing Probabilities of RNA Heterodimers. Algorithms Mol Biol. 2006, 1: 3-View ArticlePubMedPubMed CentralGoogle Scholar
- Mückstein U, Tafer H, Bernhart S, Hernandez-Rosales M, Vogel J, Stadler P, Hofacker I: Translational control by RNA-RNA interaction: Improved computation of RNA-RNA binding thermodynamics. Bioinform Res Dev. 2009, 13: 114-127.View ArticleGoogle Scholar
- Mückstein U, Tafer H, Hackermuller J, Bernhart SH, Stadler PF, Hofacker IL: Thermodynamics of RNA-RNA binding. Bioinformatics. 2006, 22: 177-1182.View ArticleGoogle Scholar
- Tafer H, Hofacker I: RNAplex: a fast tool for RNA-RNA interaction search. Bioinformatics. 2008, 24: 2657-2663.View ArticlePubMedGoogle Scholar
- Tafer H, Amman F, Eggenhofer F, Stadler PF, Hofacker IL: Fast accessibility-based prediction of RNA-RNA interactions. Bioinformatics. 2011, 27 (14): 1934-1940.View ArticlePubMedGoogle Scholar
- Kato Y, Akutsu T, Seki H: A grammatical approach to RNA-RNA interaction prediction. Pattern Recogn. 2009, 42: 531-538. 10.1016/j.patcog.2008.08.004.View ArticleGoogle Scholar
- Salari R, Backofen R, Sahinalp SC: Fast prediction of RNA-RNA interaction. Algorithms Mol Biol. 2010, 5: 5-View ArticlePubMedPubMed CentralGoogle Scholar
- Busch A, Richter AS, Backofen R: IntaRNA: efficient prediction of bacterial sRNA targets incorporating target site accessibility and seed regions. Bioinformatics. 2008, 24: 2849-2856.View ArticlePubMedPubMed CentralGoogle Scholar
- Montaseri S, Moghadam-Charkari N, Zare-Mirakabad F: A heuristic approach to RNA-RNA interaction prediction. J Theor Biol. 2012, 300: 206-211.View ArticlePubMedGoogle Scholar
- Huang FWD, Qin J, Reidys CM, Stadler PF: Target prediction and a statistical sampling algorithm for RNA-RNA interaction. Bioinformatics. 2010, 26: 175-181.View ArticlePubMedPubMed CentralGoogle Scholar
- Kato Y, Sato K, Hamada M, Watanabe Y, Asai K, Akutsu T: RactIP: fast and accurate prediction of RNA-RNA interaction using integer programming. Bioinformatics. 2010, 26: i460-i466.View ArticlePubMedPubMed CentralGoogle Scholar
- Seemann SE, Richter AS, Gesell T, Backofen R, Gorodkin J: PETcofold: predicting conserved interactions and structures of two multiple alignments of RNA sequences. Bioinformatics. 2011, 2: 211-219.View ArticleGoogle Scholar
- Salari R, Mohl M, Will S, Sahinalp SC, Backofen R: Time and space efficient RNA-RNA interaction prediction via sparse folding. 2010, RECOMB'10, Research in Computational Molecular BiologyView ArticleGoogle Scholar
- Goldberg DE: Genetic Algorithms in Search, Optimization and Machine learning, Reading. 1989, MA: Addison-WesleyGoogle Scholar
- Holland HJ: Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology. 1992, MA: Control and Artificial Intelligence. MIT Press. CambridgeGoogle Scholar
- Miettinen K, Neittaanmaki P, Periaux J: Evolutionary Programming. Genetic Programming and Industrial Applications. Evolutionary Algorithms in Engineering and Computer Science: Recent Advances in Genetic Algorithms, Evolution Strategies. 1999, New York: WileyGoogle Scholar
- Li AX, Marz M, Qin J, Reidys CM: RNA–RNA interaction prediction based on multiple sequence alignments. Bioinformatics. 2011, 4: 456-463.View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.