A novel method for prokaryotic promoter prediction based on DNA stability
© Kanhere and Bansal; licensee BioMed Central Ltd. 2005
Received: 2 September 2004
Accepted: 5 January 2005
Published: 5 January 2005
In the post-genomic era, correct gene prediction has become one of the biggest challenges in genome annotation. Improved promoter prediction methods can be one step towards developing more reliable ab initio gene prediction methods. This work presents a novel prokaryotic promoter prediction method based on DNA stability.
The promoter region is less stable and hence more prone to melting as compared to other genomic regions. Our analysis shows that a method of promoter prediction based on the differences in the stability of DNA sequences in the promoter and non-promoter region works much better compared to existing prokaryotic promoter prediction programs, which are based on sequence motif searches. At present the method works optimally for genomes such as that of Escherichia coli, which have near 50 % G+C composition and also performs satisfactorily in case of other prokaryotic promoters.
Our analysis clearly shows that the change in stability of DNA seems to provide a much better clue than usual sequence motifs, such as Pribnow box and -35 sequence, for differentiating promoter region from non-promoter regions. To a certain extent, it is more general and is likely to be applicable across organisms. Hence incorporation of such features in addition to the signature motifs can greatly improve the presently available promoter prediction programs.
Accumulation of a huge amount of genome sequence data in recent years and the task of extracting useful information from it, has given rise to many new challenges. One of the biggest challenges is the task of gene prediction and to fulfil this need, several gene prediction programs have been developed (For reviews see [1–5]). Most of these prediction programs require training based on prior knowledge of sequence features such as codon bias, which in turn are organism specific. In such cases, lack of large enough samples of known genes, as typically seen in a newly sequenced genome, can lead to sub optimal predictions. On the other hand, some gene prediction methods are based on the homology between two or more genomes but these methods are not of much help for gene prediction in case of genomes with no homologues. In addition, most of the gene prediction programs concentrate on the protein-coding regions and RNA genes, that can make up to 5 % of total protein coding genes, are neglected. Hence it is important to design ab initio gene prediction programs. One of the important steps towards ab initio gene prediction is to develop better promoter and TSS (transcription start site) prediction methods.
Although reasonable progress has been achieved in the prediction of coding region, the promoter prediction methods are still far from being accurate [6–9] and there are some very obvious reasons for these inaccuracies. One of the major difficulties is that the regulatory sequence elements in promoters are short and not fully conserved in the sequence; hence there is a high probability of finding similar sequence elements elsewhere in genomes, outside the promoter regions. This is the reason why most of the promoter prediction algorithms, which are based on finding these regulatory sequence elements, end up predicting a lot of false positives. Thus it is likely that incorporation of additional characteristics, which are unique to the promoter region, will help in improving the currently available promoter prediction methods.
In our earlier analysis, we observed that in case of bacteria as well as in eukaryotes, various properties of the region immediately upstream of TSS differ from that of downstream region . There are differences in sequence composition as well as in different sequence dependent properties such as stability, bendability and curvature. The upstream region is less stable, more rigid and more curved than downstream region. Some of these observations are supported by other studies carried out independently on genomic sequences [9, 11–17]. Among all types of promoters, the most prominent feature is the difference in DNA duplex stabilities of the upstream and downstream regions. Here, we propose a prokaryotic promoter prediction method, which is based on the stability differences between promoter and non-promoter regions.
Results and discussion
Lower stability of promoter regions in bacterial sequences
Detailed analysis of E. coli promoter sequences
Details of methodology
This difference in free energy and the stability of promoter regions as compared to that of coding and other non-coding regions can be used to search for the promoters. Based on this consideration, a new scoring function D(n) is defined, which will look for differences in free energy of the neighbouring regions of position n:
D(n) = E1(n) - E2(n)
The number of false positives obtained for different levels of sensitivity.
Cut-off for D
Cut-off for E1 (kcal/mole)
Frequency of false positives
Comparison with other promoter prediction programs
Comparison of our method with other prokaryotic prediction algorithms vis-à-vis Escherichia coli promoters.
Neural Network 
Staden's method 
Presence of high densities of promoter like signals in the upstream region of TSS may be one of the reasons why pattern matching programs result in low level of precision. This has been shown recently by a systematic analysis of sigma70 promoters from E. coli . In this study a number of weight matrices were generated by analysis of 599 experimentally verified promoters and these were tested on the 250 bp region upstream of gene start site. It was found that each 250 bp region on an average has 38 promoter-like signals. The study also presented a more rigorous patter searching method for locating promoters. With the use of this function the authors reach a sensitivity values of 0.86 but the corresponding precision achieved is only ~0.2. In case of our method, for a sensitivity of 0.9 we obtained a precision of 0.35 (as shown in Figure -9).
Recently Bockhorst et al.  proposed a very accurate method for predicting operons, promoters and terminators in E. coli. This method is based on sequence as well as expression data, but requires prior knowledge of coordinates of every ORF in the genome. We would like to emphasize here that our method is different from other methods in that it is independent of any such prior knowledge about the test gene or the organism and hence holds promise as being useful for promoter prediction in a newly sequenced genome.
The eukaryotic promoter prediction method proposed by Ohler et al.  is also worth mentioning here. Ohler et al. showed that a 30 % reduction of false positives can be achieved by use of physical properties, such as DNA bendability, in addition to other sequence properties of promoters. Interestingly, our method which also uses a physical property gives much smaller number of false positives as compared to Ohler et al.'s method. (For similar sensitivity, number of false predictions in case of Ohler et al.'s method are 1/4740 nt while in case of our method these are 1/8407 nt).
Another vertebrate promoter prediction program, 'Promfind'  identifies differences in hexanucleotide frequencies of promoter and coding region and is algorithmically quite similar to our method. But Promfind differs from our method in two important aspects. First, the Promfind program is developed mainly for vertebrate promoters and second, it assumes that in a given sequence, a promoter is always present and merely predicts its location. This need not necessarily be the case, as some of the sequences may not have any promoter at all. Our program differs from Promfind in that a promoter is predicted only when the sequence satisfies certain criteria and hence is much more appropriate for carrying out genome scale analysis.
Promoter predictions in case of RNA genes
In addition to protein coding genes there are genes present for the non-coding RNAs (ncRNAs), which play structural, regulatory and catalytic roles. It is a difficult task to find out ncRNA genes in a genome because unlike protein coding regions they lack open reading frames and also they are generally smaller in size. In addition, it is also difficult to do a homology sequence search as only the structure of ncRNA is conserved and not the sequence. There are around 156 E. coli RNA genes reported on the NCBI site  and in addition many more small RNA genes are known to exist. Argaman et al.  recently identified 14 novel sRNA genes by applying a heuristic approach to search for transcriptional signals. We have checked the performance of our algorithm with respect to the 42 RNA transcription units (TUs) reported in Ecocyc database. Our method could pick up around 57 % RNA TUs, at a cut-off corresponding to 60 % sensitivity. The program works much better in case of rRNA operons than tRNA transcription units. We could correctly pick up promoter regions in 6 out of 7 rRNA transcription units, 17 out of 33 tRNA TUs and 1 out of the 2 remaining RNA types.
Promoter prediction in Bacillus subtilis and Corynebacterium glutamicum
Finally, it is very important to see whether the method works equally well for other organisms which have genome compositions substantially different from that of Escherichia coli. Hence, we also tested our method using the promoter sequences from 1) the A+T-rich bacteria, Bacillus subtilis and 2) a G+C rich bacteria such as Corynebacterium glutamicum. Figure 9 gives a summary of the predictions in case of bacillus and corynebacterium promoters, along with those of Escherichia coli. It can be clearly seen that, at present our method performs optimally for the Escherichia coli promoters and also performs quite well in case of Bacillus subtilis. The prediction accuracy in case of Corynebacterium glutamicum promoters is not as good as that for the other two classes of promoters. However, it should be noted that the number of experimentally determined Corynebacterium promoters is much smaller as compared to other two bacteria and a larger dataset is required to arrive at any firm conclusion.
It has often been suggested that use of certain properties of promoters, other than just the sequence motifs, which can distinguish promoters from other genomic regions, could significantly improve the gene prediction methods. Although the lower stability of promoter regions as compared to non-promoter regions has been reported previously, this observation was not incorporated into a promoter prediction program. We have been able to successfully use the differential stability of promoter sequences to predict promoter regions. Our method performs better as compared to currently available prokaryotic prediction methods and is also moderately successful in predicting RNA and bacillus promoter regions. The method certainly needs to be further improved to reduce the number of predicted false positives. This can be achieved by combining the approach presented here, with the earlier reported sequence analysis methods. Such a composite method will also help in pinpointing the TSS within the promoter region identified by our method.
Promoter sequence sets
All the promoter sequences used in this study are 1000 nt long, starting 500 nt upstream (position -500) and extending up to 500 nt downstream (position +500) of the TSS. In order to avoid having multiple TSS in a given 1000 nt sequence, we have excluded all the transcription start sites which are less than 500 nt apart. Our promoter set has 227 E. coli promoters, 89 B. subtilis promoters and 28 C. glutamicum promoters.
a) Escherichia coli promoter sequences
We tested our algorithm using the Escherichia coli promoter sequences, which were taken from the PromEC dataset . The PromEC dataset provides a compilation of 471 experimentally identified transcriptional start sites. As mentioned above, after excluding all the transcription start sites which are less than 500 nt apart, the dataset contains 227 promoters. With the help of TSS information, promoter sequences were extracted from Escherichia coli genome sequence (NCBI accession no: NC_000913).
b) Bacillus subtilis promoter sequences
The transcription start sites for Bacillus subtilis promoters were obtained from the DBTBS database . The required length sequences around transcription start sites were extracted from the Bacillus genome sequence (NCBI accession no: NC_000964).
c) Corynebacterium glutamicum promoter sequences
Analysis of Corynebacterium glutamicum promoters is carried out on a set of promoters compiled by Pàtek et al.  based on experimentally determined transcription sites.
d) RNA promoter sequences
The transcription start positions of RNA transcription units are obtained from the ecocyc dataset. In this set, both computer predicted as well as experimentally determined transcription start sites, are included. In total, we have 7 rRNA TUs, 33 tRNA TUs and 2 TUs of other RNAs.
Free energy calculation
The stability of DNA molecule can be expressed in terms of free energy. The standard free energy change (ΔGo37) corresponding to the melting transition of an 'n' nucleotides (or 'n-1' dinucleotides) long DNA molecule, from double strand to single strand is calculated as follows:
ΔGoini is the initiation free energy for dinucleotide of type ij.
ΔGosym equals +0.43 kcal/mol and is applicable if the duplex is self-complementary.
ΔGoi,j is the standard free energy change for the dinucleotide of type ij.
Since our analysis involves long continuous stretches of DNA molecules, in our calculation we did not consider the two terms, ΔGoini and ΔGosym, which are more relevant for oligonucleotides. In the present calculation, each promoter sequence is divided into overlapping windows of 15 base pairs (or 14 dinucleotide steps). For each window, the free energy is calculated as given in the above equation and the energy value is assigned to the first base pair in the window. The energy values corresponding to the 10 unique dinucleotide sequences are taken from the unified parameters proposed recently [34, 35].
a) Wilcoxon signed test for equality of medians
The free energy distribution at a given position, in the 1000 nt E. coli sequences ranging from -500 to +500, was compared to the distribution in a randomly selected set. For this comparison, we followed a similar procedure as adopted by Margalit et al. . The random set was chosen such that an energy value per sequence was selected arbitrarily, independent of its position in the sequence. The comparison between the energy distributions was carried out using Wilcoxon signed test for equality of medians. This is a nonparametric test, which is used to test whether the two samples have equal medians or not.
b) Two-sample Kolmogorov-Smirnov test
We compared the free energy distribution at position -20 (with respect to TSS) with the distributions at the positions -200 and +200 using Kolmogorov-Smirnov two sample test .
All the calculations related to the statistical tests were carried out using MATLAB 6.0®.
Implementation and scoring of NNPP and Staden's method
The promoter predictions were also carried out using two other methods viz. NNPP and Staden's method. NNPP program is available at . All the NNPP predictions were carried out at a score cut-off 0.80.
In case of NNPP as well as Staden's method, the true and false positives were scored as in case of our method (Figure 3), with a prediction in -150 to 50 region being considered as a true prediction.
Sensitivity and precision
The sensitivity and precision for the predictions are calculated using the following formulae:
During the study, AK was supported by University Grants Commission and Council of Scientific and Industrial Research. We thank Prof. N. V. Joshi for his valuable comments. We also thank Dr MiroslavPátek for the Corynebacterium promoter sequences. We are grateful to the two unknown referees for their suggestions.
- Fickett JW: The gene identification problem: An overview for developers. Comput Chem 1996, 20: 103–118. 10.1016/S0097-8485(96)80012-XView ArticlePubMedGoogle Scholar
- Claverie JM: Computational methods for the identification of genes in vertebrate genomic sequences. Hum Mol Genet 1997, 6: 1735–1744. 10.1093/hmg/6.10.1735View ArticlePubMedGoogle Scholar
- Stormo GD: Gene-finding approaches for eukaryotes. Genome Res 2000, 10: 394–397. 10.1101/gr.10.4.394View ArticlePubMedGoogle Scholar
- Mathé C, Sagot MF, Schiex T, Rouzé P: Current methods of gene prediction, their strength and weaknesses. Nucleic Acids Res 2002, 30: 4103–4117. 10.1093/nar/gkf543PubMed CentralView ArticlePubMedGoogle Scholar
- Zhang MQ: Computational prediction of eukaryotic protein-coding genes. Nat Rev Genet 2002, 3: 698–709. 10.1038/nrg890View ArticlePubMedGoogle Scholar
- Fickett JW, Hatzigeorgiou AG: Eukaryotic promoter recognition. Genome Res 1997, 7: 861–78.PubMedGoogle Scholar
- Rombauts S, Florquin K, Lescot M, Marchal K, Rouze P, van de Peer Y: Computational approaches to identify promoters and cis-regulatory elements in plant genomes. Plant Physiol 2003, 132: 1162–1176. 10.1104/pp.102.017715PubMed CentralView ArticlePubMedGoogle Scholar
- Werner T: The state of the art of mammalian promoter recognition. Brief Bioinform 2003, 4: 22–30. 10.1186/1471-2105-4-22View ArticlePubMedGoogle Scholar
- Pedersen AG, Baldi P, Chauvin Y, Brunak S: The biology of eukaryotic promoter prediction – a review. Comput Chem 1999, 23: 191–207. 10.1016/S0097-8485(99)00015-7View ArticlePubMedGoogle Scholar
- Kanhere A, Bansal M: Identifcation of additional 'punctuation marks' in genomic DNA [abstract]. In proceedings of 10th congress of FAOBMB: Bangalore 139. 7–11 December 2003
- Margalit H, Shapiro BA, Nussinov R, Owens J, Jernigan RL: Helix stability in prokaryotic promoter regions. Biochemistry 1988, 27: 5179–5188. 10.1021/bi00414a035View ArticlePubMedGoogle Scholar
- Pedersen AG, Jensen LJ, Brunak S, Staerfeldt HH, Ussery DW: A DNA structural atlas for Escherichia coli. J Mol Biol 2000, 299: 907–930. 10.1006/jmbi.2000.3787View ArticlePubMedGoogle Scholar
- Choi CH, Kalosakas G, Rasmussen KO, Hiromura M, Bishop AR, Usheva A: DNA dynamically directs its own transcription initiation. Nucleic Acids Res 2004, 32: 1584–1590. 10.1093/nar/gkh335PubMed CentralView ArticlePubMedGoogle Scholar
- Levitskii VG, Katokhin AV: Computer analysis and recognition of Drosophila melanogaster gene promoters. Mol Biol (Mosk) 2001, 35: 970–978.Google Scholar
- Lisser S, Margalit H: Determination of common structural features in Escherichia coli promoters by computer analysis. Eur J Biochem 1994, 223: 823–830. 10.1111/j.1432-1033.1994.tb19058.xView ArticlePubMedGoogle Scholar
- Nakata K, Kanehisa M, Maizel JV Jr: Discriminant analysis of promoter regions in Escherichia coli sequences. Comput Appl Biosci 1988, 4: 367–71.PubMedGoogle Scholar
- Vollenweider HJ, Fiandt M, Szybalski W: A relationship between DNA helix stability and recognition sites for RNA polymerase. Science 1979, 205: 508–511. 10.1126/science.377494View ArticlePubMedGoogle Scholar
- Breslauer KJ, Frank R, Blocker H, Marky LA: Predicting DNA duplex stability from the base sequence. Proc Natl Acad Sci USA 1986, 83: 3746–3750. 10.1073/pnas.83.11.3746PubMed CentralView ArticlePubMedGoogle Scholar
- Reese MG: Application of a time-delay neural network to promoter annotation in the Drosophila melanogaster genome. Comput Chem 2001, 26: 51–56. 10.1016/S0097-8485(01)00099-7View ArticlePubMedGoogle Scholar
- Staden R: Computer methods to locate signals in nucleic acid sequences. Nucleic Acids Res 1984, 12: 505–519. 10.1093/nar/12.1Part2.505PubMed CentralView ArticlePubMedGoogle Scholar
- Mulligan ME, Hawley DK, Entriken R, McClure WR: Escherichia coli promoter sequences predict in vitro RNA polymerase selectivity. Nucleic Acids Res 1984, 12: 789–800. 10.1093/nar/12.1Part2.789PubMed CentralView ArticlePubMedGoogle Scholar
- Alexandrov NN, Mironov AA: Application of a new method of pattern recognition in DNA sequence analysis: a study of E. coli promoters. Nucleic Acids Res 1990, 18: 1847–1852. 10.1093/nar/18.7.1847PubMed CentralView ArticlePubMedGoogle Scholar
- Huerta AM, Collado-Vides J: Sigma70 promoters in Escherichia coli : specific transcription in dense regions of overlapping promoter-like signals. J Mol Biol 2003, 333: 261–278. 10.1016/j.jmb.2003.07.017View ArticlePubMedGoogle Scholar
- Burden S, Lin YX, Zhang R: Improving promoter prediction for the NNPP2.2 algorithm: a case study using E. coli DNA sequences. Bioinformatics 2004, in press.Google Scholar
- Bockhorst J, Qiu Y, Glasner J, Liu M, Blattner F, Craven M: Predicting bacterial transcription units using sequence and expression data. Bioinformatics 2003, 19(Suppl 1):i34–43. 10.1093/bioinformatics/btg1003View ArticlePubMedGoogle Scholar
- Ohler U, Niemann H, Liao G, Rubin GM: Joint modeling of DNA sequence and physical properties to improve eukaryotic promoter recognition. Bioinformatics 2001, 17(Suppl 1):S199–206.View ArticlePubMedGoogle Scholar
- Hutchinson GB: The prediction of vertebrate promoter regions using differential hexamer frequency analysis. Comput Appl Biosci 1996, 12: 391–398.PubMedGoogle Scholar
- Escherichia coli RNA genes at NCBI[http://www.ncbi.nlm.nih.gov/genomes/rnatab.cgi?gi=115&db=Genome]
- Argaman L, Hershberg R, Vogel J, Bejerano G, Wagner EG, Margalit H, Altuvia S: Novel small RNA-encoding genes in the intergenic regions of Escherichia coli . Curr Biol 2001, 11: 941–950. 10.1016/S0960-9822(01)00270-6View ArticlePubMedGoogle Scholar
- Hershberg R, Bejerano G, Santos-Zavaleta A, Margalit H: PromEC: An updated database of Escherichia coli mRNA promoters with experimentally identified transcriptional start sites. Nucleic Acids Res 2001, 29: 277. 10.1093/nar/29.1.277PubMed CentralView ArticlePubMedGoogle Scholar
- Makita Y, Nakao M, Ogasawara N, Nakai K: DBTBS: database of transcriptional regulation in Bacillus subtilis and its contribution to comparative genomics. Nucleic Acids Res 2004, 32(Database):D75–77. 10.1093/nar/gkh074PubMed CentralView ArticlePubMedGoogle Scholar
- Pàtek M, Nesvera J, Guyonvarch A, Reyes O, Leblon G: Promoters of Corynebacterium glutamicum . J Biotechnol 2003, 104: 311–323. 10.1016/S0168-1656(03)00155-XView ArticlePubMedGoogle Scholar
- SantaLucia J Jr: A unified view of polymer, dumbbell, and oligonucleotide DNA nearest-neighbor thermodynamics. Proc Natl Acad Sci USA 1998, 95: 1460–1465. 10.1073/pnas.95.4.1460PubMed CentralView ArticlePubMedGoogle Scholar
- Allawi HT, SantaLucia J Jr: Thermodynamics and NMR of internal G.T mismatches in DNA. Biochemistry 1997, 36: 10581–10594. 10.1021/bi962590cView ArticlePubMedGoogle Scholar
- Young IT: Proof without prejudice: use of the Kolmogorov-Smirnov test for the analysis of histograms from flow systems and other sources. J Histochem Cytochem 1977, 25: 935–941.View ArticlePubMedGoogle Scholar
- Hertz GZ, Stormo GD: Escherichia coli promoter sequences: analysis and prediction. Methods Enzymol 1996, 273: 30–42. full_textView ArticlePubMedGoogle Scholar
- Hertz GZ, Stormo GD: Identifying DNA and protein patterns with statistically significant alignments of multiple sequences. Bioinformatics 1999, 15: 563–577. 10.1093/bioinformatics/15.7.563View ArticlePubMedGoogle Scholar
- van Helden J: Regulatory sequence analysis tools. Nucleic Acids Res 2003, 31: 3593–3596. 10.1093/nar/gkg567PubMed CentralView ArticlePubMedGoogle Scholar
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 cited.