AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY

SEQUENCES AND APPLICATION TO INFECTIONS

DIAGNOSTIC

Victor Solovyev

, Asaf Salamov

, Igor Seledtsov

Denis Vorobyev

and Alexander Bachinsky

Department of Computer Science, Royal Holloway, University of London, London, U.K.

Softberry Inc., 116 Radio Circle, Suite 400, Mount Kisco, NY 10549, U.S.A.

Keywords: Bacterial community, Genome annotation, Sequence assembling, RNA-seq data, Computational pipelines,

classification and diagnostic pathogenic bacteria, Tanscriptome analysis.

Abstract: To annotate bacterial sequences from an environmental sample, we have developed an automatic annotation

pipeline Fgenesb_annotator that includes self-training of gene-finding parameters, prediction of CDS, RNA

genes, operons, promoters and terminators. New version of pipeline includes frame shift corrections and

special module with improved prediction accuracy of ribosomal proteins. To analyze next-generation

sequencing data we have developed OligiZip assembler and Transomics pipeline that provide solutions to

the following tasks: 1) de novo reconstruction of genomic sequence; 2) reconstruction of sequence with a

reference genome; 3) SNP discovery; 4) mapping RNA-Seq data to a reference genome, assemble them into

alternative transcripts and quantify the abundance of these transcripts. Using the OligoZip assembler and

gene Fgenesb pipeline we have developed a novel computational approach of identification toxic and non-

toxic bacterial serotypes using next-generation sequencing data. It can be used for detection of bacterial

infections in wounds, water or food contamination.

1 INTRODUCTION

Appearance of metagenomic sequencing projects

that sample the genomes of multiple species and

strains with shorter average sequence fragment

length, higher frequency of sequencing errors, and

the phylogenetic heterogeneity of the organisms in

the environmental sample, presents significant

additional challenges in computational gene finding

((Venter et al., 2004), (Mavromatis et al., 2007),

(Krause et al., 2007)). It is not surprising that some

popular gene–finding algorithms demonstrate very

poor performance on these currently abundant data

(Krause et al., 2007). To overcome this problem a

few new gene-finders have been developed such as

GISMO (Noguchi et al., 2006), which is using

Support Vector Machine for ORF classification and

MetaGene (Lowe and Eddy, 1997), which is

designed to predict genes from fragmented genomic

sequences.

The bacterial genome annotation pipeline

Fgenesb_annotator presented here was initially

developed to provide a completely automatic and

comprehensive annotation of huge amount of

sequences generated by one of the first published

metagenomic project (Venter et al., 2004). The

pipeline includes a suit of original algorithms for

protein coding gene identification, operon

assignment and promoters and terminators

identification. The parameters of gene prediction are

automatically trained using uncharacterized genomic

sequence. The pipeline also includes a module for

rRNA genes identification

and incorporates well-established public programs

such as tRNAscan-SE (Lowe and Eddy, 1997), and

Blast (Altschul et al., 1997). Fgenesb_annotator is

widely used to annotate sequences such as scaffolds

of bacterial genomes or short reads of DNA

extracted from a bacterial community ((Oliynyk et

al., 2007), (Martin et al., 2006), (Frigaard et al.,

2006), (Perez-Brocal et al., 2006), (Badger and

Olsen, 1999)).

Recent introduction of next-generation

sequencing instruments made possible sequencing

346

Solovyev V., Salamov A., Seledtsov I., Vorobyev D. and Bachinsky A..

AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY SEQUENCES AND APPLICATION TO INFECTIONS DIAGNOSTIC.

DOI: 10.5220/0003333703460353

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (Meta-2011), pages 346-353

ISBN: 978-989-8425-36-2

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

hundreds DNA of environmental sample at

dramatically reduced cost. To analyze these data we

have developed OligiZip assembler and visualization

tools that provides effective solutions to the

following three tasks: 1) ab initio reconstruction of

genomic sequence; 2) reconstruction of sequence

using a reference genome from the same or close

organism; 3) mutation profiling and SNP discovery

in a given set of genes. Using the OligoZip

assembler and our metagenomics gene prediction

pipeline FgenesB we introduce a novel

computational approach for differentiation between

toxic and non-toxic bacterial serotypes using next-

generation sequences data that can be applied for

rapid diagnostic of infections and environmental and

food contamination.

2 LEARNING PARAMETERS

AND PREDICTION OF

PROTEIN-CODING GENES

Prediction of protein-coding genes depends on a set

of parameters describing a bacterial gene model.

Many of them such as, for example, codon usage are

specific for each new genome. The Fgenesb

algorithm includes software module, which

iteratively learns these parameters using a given

genomic sequence. It starts with a compilation of all

relatively long ORFs (> 200 bp) that serve as an

initial gene set for calculating organism-specific

gene finding parameters. A few algorithms that use

long ORFs (or ORFs with a significant match to a

protein from a different organism) for deriving

models of coding sequences have been described

earlier ((Salzberg et al., 1998), (Larsen and Krogh,

2003),(Borodovsky et al., 1986). In the subsequent

iterations we assign each potential ORF a score

which reflects a joint probability to observe various

features associated with protein-coding sequences

such as coding content of a reading frame,

oligonucleotide composition of sequence regions

surrounding start and stop codons and the closeness

of ORF’s length to the length distribution of ‘real’

genes.

Protein coding sequences are modeled by 3-

periodic fifth-order Markov-chains using 5

-order

Markov transition probabilities computed for

hexamers ending at each of the three codon

positions. Non-coding sequences are also modeled

using 5

-order Markov transition matrix computed

for hexamers ending at any non-coding nucleotide.

Markov models or Hidden Markov models using

Markov transition probabilities of second or fifth-

order have earlier been implemented in many

bacterial gene-finding algorithms ((Larsen and

Krogh, 2003), (Borodovsky et al., 1986), (Krogh et

al., 1994)). Start of coding region including

ribosome-binding site (RBS) is modeled by second

order Markov model using second-order Markov

transition probabilities computed for triplets ending

at nucleotides in the upstream of the start codon

sequence [-20 - +1] (here +1 is the position of the 1

nucleotide of start codon). To derive the log-

likelihood ratios (start/nonstart probabilities) we also

computed second-order Markov non-start transition

matrix using sequences upstream of non-start triplets

ATG, GTG, TTG and TGT. These probabilities have

been averaged over windows of 5 bp long to account

for variable location of RBS in 5’-gene region. The

RBS weight matrix was applied in one of earlier

works on bacterial gene finding to improve gene

identification (Markowitz, 2007). To model the stop

codon we used second-order Markov transition

matrix computed for triplets ending at the positions

+1 to +3 after the stop codon. To derive the log-

likelihood ratios (stop/nonstop probabilities) we also

computed second-order Markov non-start transition

matrix computed for triplets ending at the positions

+1 to +3 of TAA, TAG and TGA non-stop codon

triplets. We also exploit a distribution of the ORF

lengths that is accounted in combined scoring of

potential protein-coding sequences.

If an ORF's start and end positions are L1 and L2

(excluding the stop codon), then the coding score of

the ORF (S) is defined as

S = llh(i) + llww(i)

−

∑

−

∑

+ log(P(L2 − L1+ 1) + llwm(i)

L2+ 3

∑

where llh(i) is the log-likelihood ratio for the

probability of generating nucleotide Xi in sequence

position i by Markov model of coding region to the

probability of generating nucleotide Xi by Markov

model of non-coding region computed using

corresponding Markov transition matrices of the

fifth-order described above; llww(i) is the log-

likelihood ratio for the probability of generating

nucleotide Xi in sequence position i using Markov

model of start region to the probability of generating

nucleotide Xi by Markov model of non-start region

computed using corresponding Markov transition

matrices of the second-order described above; P(L2-

L1+1) is the probability of a coding region to have

L2-L1+1 length; and llwm(i) is the log-likelihood

ratio for the probability of generating nucleotide Xi

in sequence position i using Markov model of the

stop codon region to the probability of generating

AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY SEQUENCES AND APPLICATION TO INFECTIONS

DIAGNOSTIC

347

nucleotide Xi by Markov model of non-stop region

computed using corresponding Markov transition

matrices of the second-order described above.

At each iteration step the algorithm uses gene

finding parameters produced on the previous stage

and selects ORFs with the scores higher than some

predefined threshold, sorts them in the descending

order by score and accounts any selected ORF as a

protein-coding gene if it overlaps with no more than

2 previous higher scoring genes and the fraction of

the overlap sequence does not exceed 50% of the

ORF length. Iteration stops when at least 99% of

predicted genes between two successive iterations

are the same. Optionally, we provide generic

parameters to annotate unknown bacteria, archaea or

their mixture sequences. These parameters have

been used to annotate short sequences extracted

from the environmental samples (Venter et al.,

2004).

3 ACCURACY OF PROTEIN

CODING GENE PREDICTION

BY FGENESB

Fgenesb gene prediction engine is one of the most

accurate prokaryotic gene finders. Its performance

has been estimated on various datasets. Recently the

pipeline performance has been tested on prediction

of coding sequences for the representative set of

assembled contigs (which are the typical results of

bacterial sequencing) and as well as for the

unassembled reads (Krause et al., 2007). The other

pipeline in that study was a combination of Critica

(Delcher et al., 1999) and Glimmer (Hayes and

Borodovsky, 1998) (called here CG pipeline). The

Fgenesb correctly identified 10-30% more reference

genes on the contig sequences than the CG pipeline

in every analyzed data set. The accuracy of the gene

calls on unassembled reads (where many species are

usually represented) was also evaluated. Fgenesb

correctly identified ~70% and missed ~20% of

reference genes on unassembled reads in all data

sets. The CG pipeline exhibited significantly more

poor results (~7% accurately predicted and 85%

missed genes) (Krause et al., 2007). Both pipelines

predicted 7-10% of genes inaccurately.

To make another comprehensive test we

downloaded sequences and annotations of complete

genomes from the IMG database (Besemer et al.,

2001). 216 genomes have been selected that have

percent of “unusual” annotated start codons (i.e. not

atg, gtg, tgt, ctg, att, atc, or ata) and stop codons (not

tag, taa, tga) less than 0.5%. The accuracy data for

Fgenesb, Metagene (Lowe and Eddy, 1997) and

latest versions of GeneMarkS (Delcher et al., 2007)

and Glimmer (Tyson et al., 2005) are presented in

Table 1.

Table 1: Gene prediction accuracies for 216 bacterial

complete genomes.

(%)

exact

(%)

Sn (%)

exact+overlapping

Fgenesb 76.8 91.2 96.0

GeneMarkS 73.5 90.7 96.5

Metagene* 76.8 93.7 95.9

Glimmer 73.3 88.5 95.2

*) Metagene uses a set of regression functions for

estimation the coding scores, which have been

computed using annotations of 170 bacterial

genomes (many of them are among our test set

sequences).

One more test of gene predictors we made for

artificial shotgun sequences (700 bp fragments from

a set of 216 bacterial genomes). We extracted these

sequences with real genes covering only part of each

sequence (by its 5’- or 3’-fragment) to make gene

prediction more difficult. Correct gene assignment

for these sequences was extremely high for all three

gene finders (Table 2), while Metagene shows

slightly lower specificity than Fgenesb and

GeneMark.

Table 2: Gene prediction accuracies for short

bacterial sequences.

(Sn+Sp)/2 (%) Sn (%) Sp (%)

Fgenesb 95.55 98.5 93.5

GeneMark 94.05 98.8 91.3

Metagene 91.65 97.6 82.9

Our results presented in Table 1 and 2 are in

close agreement with the performance analysis of

Metagene and GenMark tested on different data sets

earlier (Lowe and Eddy., 1997).

The final Fgenesb_annotator results can be

presented in the GenBank format to be readable by

visualization software such as Softberry Bacterial

Genome Explorer by server or local versions (Figure

1) or converted into the format that is required by

SEQIN software for submission of new sequences

with their annotations to the GenBank.

The Fgenesb_annotator has been applied in

dozens of published studies for annotation of whole

bacterial genomes, as well as short sequences

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

348

Figure 1: Bacterial Genome Explorer. Annotation of a set of matagenomics sequences is visualized.

Figure 2: OligoZip. de novo assembling algorithm.

extracted form environmental samples (for examples,

((Martinez et al., 2007), (McClain et al., 2009), (Yan

et al., 2004)). The operons identified by

Fgenesb_annotator have been used as initial operon

models that were further experimentally investigated

and improved ((Yan et al., 2006), (Pothier et al.,

2007)). Other pipeline components such as Bprom

(promoter predictor) and Bterm (terminator

predictor) also have been actively used in many

studies of functional characterization of bacterial

sequences ((Budde et al., 2007), (Pilhofer et al.,

2007), (Kosaka et al., 2006), (Grieshaber et al.,

2006), (Warren et al., 2007)).

4 OLIGOZIP TOOLS FOR

RECONSTRUCTING

SEQUENCES FROM

NEXT-GENERATION

SEQUENCING DATA

Recent introduction of next-generation sequencing

instruments: 454 (Roche), Genome Analyzer

AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY SEQUENCES AND APPLICATION TO INFECTIONS

DIAGNOSTIC

349

Figure 3: Sequence Explorer to visualizing contigs and assembled reads.

(Illumina), ABI-SOLiD (Life Technologies), and

several others, all capable of producing millions of

DNA sequence reads in a single run, made possible

sequencing hundreds of genes in parallel at

dramatically reduced cost. We have developed a

new algorithm (Fig.2) for processing short reads

from next-generation sequencing machines.

OligoZip tool uses L-plets hashing technique to

achieve fast data processing, and it takes into

account reads quality information.

OligoZip provides effective solutions the

following tasks: 1) De novo reconstruction of

genomic sequence; 2) Reconstruction of sequences

based on reference genome from same or close

species; 3)Mutation profiling and SNP discovery in

a given set of genes.

De novo sequence reconstruction was tested on

assembling several phage and bacterial genomes

demonstrated its superior clustering power compared

to earlier published approach [35]. Simulated error-

free 25mers of bacteriophage φX174 and

coronavirus SARS TOR2 were assembled perfectly

and in reconstruction of Haemophilus influenzae

genome contigs assembled by OligoZip were almost

twice as long as those assembled by the published

SSAKE software.

To test reconstruction of bacterial sequences

using reference genome, we assembled genomic

sequence of Methanopyrus kandleri TAG11 on

known genome of Methanopyrus kandleri AV19.

Solexa reads, about 6 million each for AV19 and

TAG 11, were produced by sequencing lab of

Harvard Partners HealthCare Center for Genetics

and Genomics. AV19 genome itself has been

assembled perfectly, with one extra contig that

happened to be the genome of phage φX174. TAG

11 reads were assembled into several hundred

contigs. Similar results were achieved on five other

Methanopyrus stains. We also develop sequence

assembling viewer to work with the reads data and

assembling results interactively (Fig 3).

5 DIFFERENTIATION BETWEEN

TOXIC AND NON-TOXIC

BACTERIAL SEROTYPES

Using the OligoZip assembler and our

metagenomics gene prediction pipeline FgenesB we

have developed a novel computational approach of

differentiation between toxic and non-toxic bacterial

serotypes using next-generation sequences data.

Initially we found a way to construct a plot that

separates well non-pathogenic and pathogenic

genomes by using comparison of the bacterial

proteomes predicted by Fgenesb (fig. 4).

Further, we have modeled next generation

sequencing data extracted from the environmental

sample. We generated a set of oligs 200 bp long for

a mixture of 10 genomes. After that we tested how

accurate a pathogenic genome can be identified in

the mixture by aligning all oligs to a set of known

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

350

Table 3: Alignment of a genome oligs to the same or other genomes.

enome

Oligs from

Gallinarum

(nonpathogenic)

Typhi_CT18

(pathogenic)

Bacillus subtilis

168

Hits Total

Score

Hits Total

Score

Hits Total

Score

allinarum

(

nonpathogenic)

215622

206264

6407000

6340000

208461

192903

5946000

5837000

67930

15278

591066

234807

yphi_CT18

(

pathogenic)

210048

198050

6171000

6087000

226918

219868

7406000

7355000

70377

16126

615816

248026

Figure 4: The plot of comparison of 32 known complete

E.coli. genomes. All non-pathogenic genomes are

cauterized in the circled region.

genomes. A typical example of alignment scores is

presented in table 3 where significant alignments

(hits) is counted to the same and different genomes.

We computed the total number of hits for 2 score

thresholds as well as total score of these hits. We can

observe from the table that if a pathogenic genome is

present in the mixture it will be recognized, while

some bacterial strains have very similar genomes.

We are observing many common oligs between

close pathogenic and non-pathogenic genomes. To

increase sensitivity of classification of genomes

from environmental sample we applied OligoZip for

assembling a set of generated oligs from the genome

mixture. OligoZip has produced ~ 100 contigs for

each genome (with using 10 times coverage) or ~ 50

contigs for each genome (using 20 times coverage).

These contigs cover ~ 99% of genome sequences.

The contigs sizes range from 755000 bases to 2000

bp. Using sequences of these contigs we can identify

the genomes from the mixture with no ambiguity.

Annotating these sequences by Fgenesb gene-finder

we can compare proteome content of known

genomes with proteins in the mixture contigs and

localize the mixture genomes on the plot of

pathogenic and non-pathogenic genomes.

Currently we are collaborating with Veterinary

Laboratories Agency to receive next generation

sequencing data of environmental samples. The

describe approach can be applied for analyzing such

DNA for detection of bacterial toxic and non-toxic

serotypes in wounds, water or food contaminations.

In addition we implemented Transomics

computational pipeline to work with next generation

transcriptome data. The pipeline maps RNA-Seq

data to a reference genome, assemble them into

transcripts and quantify the abundance of these

transcripts in particular datasets. The pipeline will

include the following key components: alignment of

RNA-Seq reads to genome, identification of

alternative transcripts, (for eukaryotic genes) and

measuring expression levels of transcript isoforms.

This pipeline also can be applied to analysis of

transcriptomics data to detect pathogenic viruses or

bacteria.

Fgenesb, OligoZip or Transomics pipeline

components and other software programs are

available to run independently at

www.softwberry.com or as a part of integrated

environment of the Molquest software package that

can be downloaded at www.molquest.com for

Windows, MAC and Linux OS.

REFERENCES

Venter, J. C., Remington, K., Heidelberg, J., Halpern, A.,

Rusch, D., Eisen, J., Wu, D., Paulsen, I., Nelson, K.,

Nelson, W., et al. (2004) Environmental genome

shotgun sequencing of the Sargasso Sea. Science, 304,

66–74.

Mavromatis, K., Ivanova, N., Barry, K., Shapiro, H.,

Goltsman,E., McHardy, A. C., Rigoutsos, I., Salamov,

A., Korzeniewski, F., Land, M., et al. (2007) Use of

simulated data sets to evaluate the fidelity of

metagenomic processing methods. Nature Methods, 4,

495-500.

Krause, L., McHardy, A., Nattkemper, T., Puhler, A.,

Stoye, J., Meyer, F. (2007) GISMO - gene

identification using a support vector machine for ORF

classification. Nucleic Acids Res., 35, 2,

540-549.

Noguchi, H., Park, J., Takagi, T. (2006) MetaGene:

prokaryotic gene finding from environmental genome

shotgun sequences. Nucleic Acids Res., 34, 19, 5623–

5630.

AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY SEQUENCES AND APPLICATION TO INFECTIONS

DIAGNOSTIC

351

Lowe, T. M. and Eddy, S.R. (1997) tRNAscan-SE: a

program for improved detection of transfer RNA

genes in genomic sequence. Nucleic Acids Res., 25,

955-964.

Altschul, S. F., Madden, T. L., Schaffer, A. A., Zhang, J.,

Zhang, Z., Miller, W., Lipman, D. J. (1997) Gapped

BLAST and PSI-BLAST: a new generation of protein

database search programs. Nucleic Acids Res., 25,

3389-3402.

Oliynyk, M., Samborskyy, M., Lester, J., Mironenko, T.,

Scott, N., Dickens, S., Haydock, S., Leadlay, P. (2007)

Complete genome sequence of the erythromycin-

producing bacterium Saccharopolyspora erythraea

NRRL23338. Nature Biotechnology, 25, 447–453.

Martin, H., Ivanova, N., Kunin,V., Warnecke, F., Barry,

K., McHardy, A., Yeates, C., He, S., Salamov, A.,

Szeto, S., et al. (2006) Metagenomic analysis of two

enhanced biological phosphorus removal (EBPR)

sludge communities. Nature Biotechnology, 24, 1263–

1269.

Frigaard, N., Martinez, A., Mincer, T., DeLong, E. (2006)

Proteorhodopsin lateral gene transfer between marine

planktonic Bacteria and Archaea. Nature, 439, 847-

850.

Perez-Brocal,V., Gil, R., Ramos, S., Lamelas, A., Postigo,

M., Michelena, J., Silva, F., Moya, A., Latorre, A.

(2006) A Small Microbial Genome: The End of a

Long Symbiotic Relationship? Science, 314, 312-313.

Badger, J. H., Olsen, G. J. (1999) CRITICA: coding

region identification tool invoking comparative

analysis. Mol. Biol. Evol., 16, 512–524.

Delcher, A., Harmon, D., Kasif, S., White, O., Salzberg, S.

(1999) Improved microbial gene identification with

GLIMMER. Nucleic Acids Res., 27, 4636–4641.

Hayes,W., Borodovsky, M. (1998) How to interpret an

anonymous bacterial genome: machine learning

approach to gene identification. Genome Res., 8, 1154-

1171.

Salzberg, S., Delcher, A., Kasif, S., White, O. (1998)

Microbial gene identification using interpolated

Markov models. Nucleic Acids Res., 26, 544-548.

Larsen, T., Krogh, A. (2003) EasyGene – a prokaryotic

gene finder that ranks ORFs by statistical significance.

BMC Bioinformatics, 4, 21, 1-15.

Borodovsky, M. Y., Sprizhitskii, Y. A., Golovanov, E. I.,

Aleksandrov, A. A. (1986) Statistical patterns in

primary structures of functional regions in E. coli

genome: III. Computer recognition of coding regions.

Mol. Biol., 20, 1390-1398.

Krogh, A., Mian, I., Haussler, D. (1994) A hidden Markov

model that finds genes in E. coli DNA. Nucleic Acids

Res., 22, 4768–4778.

Frishman, D., Mironov, A., Mewes, H. W., Gelfand, M.

(1998) Combining diverse evidence for gene

recognition in completely sequenced bacterial

genomes. Nucleic Acids Res., 26, 2941-2947.

Markowitz, V. M., Microbial genome data sources. Curr.

Opin. Biotechnol. 18, 267–272 (2007).

Besemer, J., Lomsadze, A., Borodovsky, M. (2001)

GeneMarkS: a self-training method for prediction of

gene starts in microbial genomes. Implications for

finding sequence motifs in regulatory regions. Nucleic

Acids Res., 29, 2607-2618.

Delcher, A, Bratke, K., Powers, E., Salzberg, S. (2007)

Identifying bacterial genes and endosymbiont DNA with

Glimmer. Bioinformatics 23, 6, 673–679.

Tyson, G., Lo, I., Baker,B., Allen, E., Hugenholtz, P.,

Banfield, J. (2005) Genome-directed isolation of the

key nitrogen fixer Leptospirillum ferrodiazotrophum

sp. nov. from an acidophilic microbial community.

Appl. Envir. Microbiol., 71, 6319-6324.

Martinez, A., Bradley, A. S., Waldbauer, J. R., Summons,

R. E., and DeLong, E. F. (2007) Proteorhodopsin

photosystem gene expression enables

photophosphorylation in a heterologous host. PNAS,

104, 5590-5595.

McClain, M., Shaffer, C., Israel, D., Peek, R., Cover, T.

(2009) Genome sequence analysis of Helicobacter

pylori strains associated with gastric ulceration and

gastric cancer. BMC Genomics, 10, 3.

Yan, B., Methé, B. A., Lovley, D. R., Krushkal, J. (2004)

Computational prediction of conserved operons and

phylogenetic footprinting of transcription regulatory

elements in the metal-reducing bacterial family

Geobacteraceae. J. Theor. Biol., 230, 133-144.

Yan, B., Núñez, C., Ueki, T., Esteve-Núñez, A., Puljic.M.,

Adkins, R. M., Methé, B. A., Lovley, D. R., Krushkal,

J. (2006). Computational prediction of RpoS and

RpoD regulatory sites in Geobacter sulfurreducens

using sequence and gene expression information.

Gene, 384, 73-95.

Pothier, J. F., Wisniewski-Dye, F., Weiss-Gayet, M.,

Loccoz, Y., Prigent-Combaret, C. (2007) Promoter-

trap identification of wheat seed extract-induced genes

in the plant-growth-promoting rhizobacterium

Azospirillum brasilense Sp245. Microbiology, 153,

3608–3622.

Gil, H., Platz, G. J., Forestal, C. A., Monfett, M., Bakshi,

C., Sellati, T. J., Furie. M. B., Benach, J. L., Thanassi,

D. G. (2006) Deletion of TolC orthologs in Francisella

tularensis identifies roles in multidrug resistance and

virulence. PNAS, 103, 12897-12902.

Michel, G. P., Durand, E., Filloux, A. (2007)

XphA/XqhA, a Novel GspCD Subunit for Type II

Secretion in Pseudomonas aeruginosa. J. Bacteriol.,

189, 3776-3783.

Budde, P., Davis, B., Yuan, J., Waldor, M. (2007)

Characterization of a higBA Toxin-Antitoxin Locus in

Vibrio cholerae. J. Bacteriol., 189, 491-500.

Pilhofer, M., Bauer, A., Schrallhammer, M., Richter, L.,

Ludwig, W., Schleifer, K., Petroni, G. (2007)

Characterization of bacterial operons consisting of two

tubulins and a kinesin-like gene by the novel Two-

Step Gene Walking method. Nucleic Acids Res., 35,

e135.

Kosaka, T., Uchiyama, T., Ishii, S., Enoki, M., Imachi, H.,

Kamagata, Y., Ohashi, A., Harada, H., Ikenaga, H., and

Watanabe, K. (2006) Reconstruction and Regulation

of the Central Catabolic Pathway in the Thermophilic

Propionate-Oxidizing Syntroph Pelotomaculum

thermopropionicum. J. Bacteriol., 188, 202-210.

Grieshaber, N. A., Grieshaber, S. S., Fischer, E. R.,

Hackstadt, T. (2006) A small RNA inhibits translation

BIOINFORMATICS 2011 - International Conference on Bioinformatics Models, Methods and Algorithms

352

of the histone-like protein Hc1 in Chlamydia

trachomatis. Mol. Microbiol., 59, 541-550.

Warren

R., Sutton G, Jones S and Holt R. Assembling

millions of short DNA sequences using SSAKE.

Bioinformatics, 2007, 23(4), 500-501.

AUTOMATIC ANNOTATION OF BACTERIAL COMMUNITY SEQUENCES AND APPLICATION TO INFECTIONS

DIAGNOSTIC

353