A Pipeline for Insertion Sequence Detection and Study for Bacterial Genome

TL;DR

This paper presents a new pipeline combining existing tools for accurate detection and classification of insertion sequences in bacterial genomes, tested on Pseudomonas aeruginosa genomes.

Contribution

The authors developed a comprehensive pipeline integrating multiple annotation tools and IS detection methods, improving accuracy and enabling detailed IS analysis in bacterial genomes.

Findings

Prodigal outperforms other annotation tools for IS prediction.

Close genomes in P. aeruginosa have similar IS family distributions.

The pipeline effectively detects and classifies IS elements in bacterial genomes.

Abstract

Insertion Sequences (ISs) are small DNA segments that have the ability of moving themselves into genomes. These types of mobile genetic elements (MGEs) seem to play an essential role in genomes rearrangements and evolution of prokaryotic genomes, but the tools that deal with discovering ISs in an efficient and accurate way are still too few and not totally precise. Two main factors have big effects on IS discovery, namely: genes annotation and functionality prediction. Indeed, some specific genes called "transposases" are enzymes that are responsible of the production and catalysis for such transposition, but there is currently no fully accurate method that could decide whether a given predicted gene is either a real transposase or not. This is why authors of this article aim at designing a novel pipeline for ISs detection and classification, which embeds the most recently available tools developed in this field of research, namely OASIS (Optimized Annotation System for Insertion Sequence) and ISFinder database (an up-to-date and accurate repository of known insertion sequences). As this latter depend on predicted coding sequences, the proposed pipeline will encompass too various kinds of bacterial genes annotation tools (that is, Prokka, BASys, and Prodigal). A complete IS detection and classification pipeline is then proposed and tested on a set of 23 complete genomes of Pseudomonas aeruginosa. This pipeline can also be used as an investigator of annotation tools performance, which has led us to conclude that Prodigal is the best software for IS prediction. A deepen study regarding IS elements in P.aeruginosa has then been conducted, leading to the conclusion that close genomes inside this species have also a close numbers of IS families and groups.

Figures

Tables

Table 1. Table 1: Input set of 23 complete genomes of P. aeruginosa

		INSDC(Genbank)	Refseqs	Input Cenomes
Index	GenomeName	GID	GID	GID	Accession no.
1	PACS2	106896550	_	106896550	AAQW01000001.1
2	PAO1	110227054	110645304	110645304	NC_002516.2
3	UCBPP-PA14	115583796	116048575	116048575	NC_008463.1
4	PA7	150958624	152983466	152983466	NC_009656.1
5	19BR	343788106	485462089	485462089	NZ_AFXJ01000001.1
6	213BR	343788107	485462091	485462091	NZ_AFXK01000001.1
7	M18	347302377	386056071	386056071	NC_017549.1
8	DK2	392316915	392981410	392981410	NC_018080.1
9	B136-33	477548288	478476202	478476202	NC_020912.1
10	RP73	514245605	514407635	514407635	NC_021577.1
11	c7447m	543873856	_	543873856	CP006728
12	PAO581	543879514	_	543879514	CP006705
13	PAO1-VE2	553886202	_	553886202	CP006831
14	PAO1-VE13	553895034	_	553895034	CP006832
15	PA1	557703951	558672313	558672313	NC_022808.1
16	PA1R	557709751	558665962	558665962	NC_022806.1
17	MTB-1	563408818	564949884	564949884	NC_023019.1
18	LES431	566561164	568151185	568151185	NC_023066
19	SCV20265	567363169	568306739	568306739	NC_023149
20	LESB58	_	218888746	218888746	NC_011770.1
21	NCGM2.S1	_	386062973	386062973	NC_017549.1
22	PA38182	575870901	_	575870901	HG530068.1
23	YL84	576902775	_	576902775	CP007147.1

Table 2. Table 2: Summary table produced by NOASIS (begining)

Name	Genome	Start	End	Orientation	SetID	ISFinder_name	Family	Group	Length
PAO1	NC_002516.2	499832	501193	-	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	2556875	2558236	+	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	3043478	3044839	-	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	3842002	3843363	-	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	4473550	4474911	+	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	5382524	5383885	-	1	ISPa11	IS110	IS1111	1361
PAO1	NC_002516.2	54041	54835	+	2	ISStma5	IS3	IS3	794

Table 3. Table 3: Summary table produced by NOASIS (end)

IRR=IRL	Locus_tag(gbk)	Product(gbk)	E Value	IS_type
ATGGACTCCTCCC	[[’PA0445’]]	[[’transposase’]]	0.0	Real IS
ATGGACTCCTCCC	[[’PA2319’]]	[[’transposase’]]	0.0	Real IS
ATGGACTCCTCCC	[[’PA2690’]]	[[’transposase’]]	0.0	Real IS
ATGGACTCCTCCC	[[’PA3434’]]	[[’transposase’]]	0.0	Real IS
ATGGACTCCTCCC	[[’PA3993’]]	[[’transposase’]]	0.0	Real IS
ATGGACTCCTCCC	[[’PA4797’]]	[[’transposase’]]	0.0	Real IS
AAAGGGGACAGATTTATTTTCCCTGCTCTAAT	[[’PA0041a’]]	[[’transposase’]]	0.23	Putative New IS

Table 4. Table 4: BASys annotation using NOASIS and DOASIS

BASys		Normal	All Transpos	Zigzag Odd	Zigzag Even	(All_T/odd/even)
Name	Genome	Real IS	Real IS	Real IS	Real IS	Best Real
PACS2	106896550	2	1	2	0	2
PAO1	110645304	9	6	0	0	6
UCBPP-PA14	116048575	3	8	8	1	8
PA7	152983466	13	0	0	0	0
LESB58	218888746	2	3	5	2	6
M18	386056071	1	2	2	1	2
NCGM2.S1	386062973	15	0	12	0	12
DK2	392981410	8	9	10	8	11
B136-33	478476202	3	5	3	3	5
19BR	485462089	5	0	0	10	10
213BR	485462091	5	4	4	4	4
RP73	514407635	4	5	5	2	5
c7447m	543873856	9	0	9	9	10
PAO581	543879514	9	6	8	0	8
PAO1-VE2	553886202	8	6	9	6	9
PAO1-VE13	553895034	8	6	8	8	8
PA1R	558665962	4	4	4	5	5
PA1	558672313	5	5	5	6	6
MTB-1	564949884	0	1	0	1	1
LES431	568151185	5	14	13	8	14
SCV20265	568306739	5	14	13	8	14
PA38182	575870901	1	3	1	2	4
YL84	576902775	7	7	7	7	7
		131	109	128	91	157

Table 5. Table 5: Correlation table for different annotation tools

	NCBI		BASys		Prokka
	NOASIS	DOASIS	NOASIS	DOASIS	NOASIS	DOASIS
Number of Real IS	110	169	131	157	169	176
Correlations	0.3985580752		0.357346472		0.926355615

Table 6. Table 6: Final comparison using our pipeline

	NCBI	BASys	Prokka	Prodigal		NCBI	BASys	Prokka	Prodigal
Name	DOASIS	DOASIS	DOASIS	DOASIS	Name	DOASIS	DOASIS	DOASIS	DOASIS
PACS2	0	2	2	3	c7447m	10	10	10	12
PAO1	10	6	10	12	PAO581	10	8	10	8
UCBPP-PA14	4	8	4	8	PAO1-VE2	10	9	10	9
PA7	15	0	14	18	PAO1-VE13	10	8	10	9
LESB58	3	6	3	6	PA1R	4	5	4	5
M18	3	2	3	3	PA1	5	6	5	6
NCGM2.S1	11	12	19	14	MTB-1	1	1	1	2
DK2	12	11	13	17	LES431	6	14	6	6
B136-33	5	5	5	7	SCV20265	15	14	15	15
19BR	8	10	5	11	PA38182	7	4	7	7
213BR	8	4	8	10	YL84	7	7	7	8
RP73	5	5	5	5	Total IS	84	71	91	114

Full text

1 Introduction

The number of completely sequenced bacterial and archaeal genomes are rising steadily, such an increasing makes it possible to develop novel kind of large scale approaches to understand genomes structure and evolution over time. Gene content prediction and genome comparison have both provided new important information and deciphering keys to understand evolution of prokaryotes  [14]. Important sequences in understanding rearrangement of genomes during evolution are so-called transposable elements (TEs), which are DNA fragments or segments that have the ability to insert themselves into new chromosomal locations, and often make duplicate copies of themselves during transposition process [5]. Remark that, in bacterial reign, only cut-and-paste mechanism of transposition can be found, the transposable elements involved in such a move being the insertion sequences (ISs).

Insertion sequences range in size from 600 to more than 3000 bp. They are divided into 26 main different families in prokaryotes, as described in ISFinder111www-is.biotoul.fr [12], an international reference database for bacterial and archaeal ISs that includes background information on transposons. The main function of ISFinder is to assign IS names and to produce a focal point for a coherent nomenclature for all discovered insertion sequences. This database includes over than 3500 bacterial ISs [6, 17]. Data come from a detection of repeated patterns, which can be easily found by using homology-based techniques [3]. Classification process of families, for its part, depends on transposases homology and overall genetic organization. Indeed, most ISs consist of short inverted repeat sequences that flank one or more open reading frames (ORFs, see Figure 1), whose products encode the transposase proteins necessary for transposition process. The main problem with such approaches for ISs detection and classification is that they are obviously highly dependent on the annotations, and existing tools evoked above only use the NCBI ones, whose quality is limited and very variable.

In this research work, the authors’ intention is to find an accurate method for discovering insertion sequences in prokaryotic genomes. To achieve this goal, we propose to use one of the most recent computational tool for automated annotation of insertion sequences, namely OASIS, together with the international database for all known IS sequences (ISFinder). More precisely, OASIS works with genbank files that have fully described genes functionality: this tool identifies ISs in each genome by finding conserved regions surrounding already-annotated transposases. Such technique makes it possible to discover new insertion sequences, even if they are not in ISFinder database. A novel pipeline that solves the dependence on NCBI annotations, and that works with any annotation tool (with or without description of gene functionality) is then proposed. The output of our pipeline contains all detected IS sequences supported with other important information like inverted repeats (IRs) sequences, lengths, positions, names of family and group, and other details that help in studying IS structures.

The contributions of this article can be summarized as follows. (1) A pipeline for insertion sequences discovery and classification is proposed, which does not depend on NCBI annotations. It uses unannotated genomes and embeds various annotation tools specific to Bacteria (such as Prokka, BASys, and Prodigal) in its process. (2) Overlapping and consensus problems that naturally appear after merging annotation methods recalled above are solved, in order to obtain large and accurate number of ISs with their names of families and groups. And finally (3) the pipeline is tested on a set of 23 complete genomes of Pseudomonas aeruginosa, and biological consequences are outlined.

The remainder of this article is organized as follows. In Section 2, various tools for discovering IS elements in different species of Bacteria and Archaea are presented. The suggested methodology for increasing both the number and accuracy of detecting IS elements is explained in Section 3. The pipeline is detailed in Section 4, while an application example using 23 completed genomes of P. aeurigonsa is provided in Section 5. This article ends by a conclusion section, in which the contributions are summarized and intended future work is detailed.

2 State of the art in ISs detection or annotation

The study on the plant-pathogenic prokaryote Xanthomonas oryzae pv. oryzae (Xoo), which causes bacterial blight (one of the most important diseases of rice) was published in 2005 by Ochiai et al. [8]. They used GeneHacker [16], GenomeGambler version 1.51, and Glimmer program [2] for coding sequence prediction. Insertion sequences were finally classified by a BLAST analysis using ISFinder database evoked previously.

IScan, developed by Wagner et al. [15], has then been proposed in 2007. Inverted repeats are found using smith waterman local alignments on transposase references found with BLAST and used as a local database. This tool has been applied on 438 completely sequenced bacterial genomes by using BLAST with referenced transposases, to determine which transposases are related to insertion sequences. Touchon et al., for their parts, have analyzed 262 different bacterial and archaeal genomes downloaded from GenBank NCBI in 2007 [13]. A coding sequence has then been considered as an IS element if its BLASTP best hit in ISFinder database has an e-value lower than $10^{-10}$.

ISA has been created by Zhou et al. in 2008 [17]. This annotation program depends on both NCBI annotations and ISFinder. More precisely, authors manually collected 1,356 IS elements with both sequences and terminal signals from the ISFinder database, which have been used as templates for identification of all IS elements and map construction in the targeted genomes. ISA, which is not publicly available, has finally been used for an analysis of 19 cyanobacterial and 31 archaeal annotated genomes downloaded from NCBI.

In 2010, Plague et al. analyzed the neighboring gene orientations (NGOs) of all ISs in 326 fully sequenced bacterial chromosomes. They obtained primary annotations from the Comprehensive Microbial Resource database (release 1.0-20.0) at the Institute for Genomic Research222http://cmr.tigr.org/tigr-scripts/CMR/CmrHomePage.cgi. Their approach for extracting IS elements from these genomes was to consider that a coding sequence with a best BLASTX hit e-value lower than $10^{-10}$ is an insertion sequence [9]. ISsage, for its part, has been developed in 2011 by Varani et al. [14]. They used eight different bacterial genomes downloaded from NCBI, and produced a web application pipeline that allows semi-automated annotation based on BLAST against the ISFinder database. However ISsage cannot automatically identify new insertion sequences which are not already present in ISFinder database.

A new computational tool for automated annotation of ISs has then been released in 2012 by Robinson et al. [10]. This tool has been called OASIS, which stands for “Optimized Annotation System for Insertion Sequences”. They worked with 1,737 bacterial and archaeal genomes downloaded from NCBI. OASIS identifies ISs in each genome by finding conserved regions surrounding already-annotated transposase genes. OASIS uses a maximum likelihood algorithm to determine the edges of multicopy ISs based on conservation between their surrounding regions. For defining inverted repeats, the same strategy as IScan was used (Smith-Waterman alignment). Authors also used hierarchical agglomerative clustering to identify groups of IS lengths. The ISs set is then classified according to the family and group after a BLASTP best hit in ISFinder database with an e-value lower than $10^{-12}$. When a cluster cannot match with any entry of the database, the IS set is considered as new. Thus OASIS has the ability to discover new insertion sequences, that is, which cannot be found in ISFinder.

Finally, in 2014, the analysis of the NGOs for all IS elements within 155 fully sequenced Archaea genomes was presented by Florek et al. [4]. To do so, they have launched a BLASTP in the ISFinder, with an e-value less than or equal to $10^{-10}$, for all protein coding sequences downloaded from NCBI which are related to IS elements.

Two major concerns with the state of the art detailed above can be emphasized. Firstly, most of them cannot detect new insertion sequences. Secondly, all these tools are based on NCBI annotations of very relative and variable qualities – except ISsaga, which could work with other annotation tools (but it depends only on transposase ORFs that have been already defined in ISFinder). Our objective in the next section is to propose a pipeline that solves these two issues, being able to deal with unannotated genomes and to detect unknown ISs.

3 Prediction and Modules based on OASIS

For illustration purpose, the proposed pipeline system for IS elements prediction will be presented using 23 complete genomes of P. aeruginosa available on the NCBI website, RefSeq and INCDS/Genebank databases, see Table 1 (RefSeq genomes were prefered when available). The prediction of IS elements in the proposed pipeline depends on both OASIS [10] and ISFinder [12].

3.1 Prediction of IS elements from Pseudomonas aeruginosa

OASIS is used in this pipeline for predicting insertion sequences in prokaryotic genomes. This latter detects ISs in each genome by finding conserved regions surrounding already-annotated transposase genes, which are identified by the word $\textasciidieresis$transposase$\textasciidieresis$ in the “product” field of the GenBank file. Obviously OASIS highly depends on the quality of annotations [10], while to determine whether a given gene is a transposase or not is a very difficult task (indeed transposases are among the most abundant and ubiquitous genes in nature [1], and they are widely separated in Prokaryote genomes). OASIS deals with files having genbank format. It takes them as input and then produces two output files for each provided genome. The first one is a fasta file that contains all IS nucleotide sequences, with start and end positions. It also contains the amino acid sequence for each ORF. The second file is a summary table providing attributes that describe the insertion sequence: set-id, family, group, IS positions, inverted repeat left (IRL) and right (IRR), and orientation. Remark that most of these information are in the ISFinder database too. Indeed OASIS find them alone but it extracts family names and group from ISFinder.

The main problem found in OASIS is solved in the proposed pipeline by using different types of annotations: NCBI will not be used alone, and gene functionality taken from annotation tools will either or not be used depending on the situation. Finally, transposases within IS will be verified using ISFinder database. OASIS can thus be used in two different ways in our pipeline, depending on the provided genbank file. These two modules have been named NOASIS, which uses the original input genbank genome file provided by the NCBI (as it is, without any modification), and DOASIS, which deals with modified genbank files that have been updated to obtain more accurate results than NOASIS. These modules are described thereafter.

3.2 Normal OASIS (NOASIS)

For finding predicted IS in NOASIS module, we simply applied OASIS on the input set of genomes with their NCBI annotations, that is, with the original downloaded genbank file. Using the reference genome named PAO1, the summary outputted by the pipeline is given in Tables 2 and 3. In these NOASIS tables, the summary produced by OASIS is enriched with new features described below:

• •

Real IS IS sequences that have best match (first hit) when using BLASTN with ISFinder database, an e-value equal to 0.0, and with a functionality of each ORF within the IS recognized as a transposase.

• •

Partial IS Sequences that match part of known IS from ISFinder (i.e., have e-value lower than $10^{-10}$) and have also a transposase gene functionality for the ORFs.

• •

Putative New IS Sequences with bad score after making a BLASTN with ISFinder, but with a transposase. They may be real insertion sequences not already added in ISFinder database or false positives, requiring human curation.

Applying this slightly improved version of OASIS in the 23 genomes of Pseudomonas leads to a major issue: surprisingly, NOASIS found no real insertion sequences in some genomes like PACS2 or SCV20265. The problem is that OASIS find multiple copies of IS elements in each genome by identifying conserved regions surrounding transposase genes. However some of the considered genomes either have no information about transposase gene into their feature genbank tables or have simply no feature table in their genbank format files. This issue is at the basis of our improved module called DOASIS, which is explained below.

For the sake of comparison, Figure 2 contains similar results for Mycobacterium tuberculosis genus.

3.3 Developed OASIS (DOASIS)

The main idea for DOASIS module is that information about transposases within genbank files are potentially incorrect (i.e., may all be false positives). So we simply decide to remove all transposase words in the product fields from all inputted genomes. We thus update these information as follows.

Step 1: genbank update.

Inputted genbank files are modified following one of the three methods below.

1. 1.

All-Tpase: we consider that all the genes may potentially be a transposase. So all product fields are set to “transposase”. 2. 2.

Zigzag Odd: we suggest that genes in odd positions are putative transposases and we update the genbank file adequately. Oddly, this new path will produce new candidates which are not detected during All-Tpase. 3. 3.

Zigzag Even: similar to Zigzag Odd, but on even positions.

We checked also a randomized method (i.e., by putting “transposase” in randomly picked genes). However we found poorer number of predictive real ISs or new real ISs compared with the three methods previously presented. For these reasons, we will not further investigate the randomized method.

Step 2.

We apply OASIS three times (i.e., one time per method) on all genomes, and then we take the output fasta file that contains both nucleotides and amino acids sequences for each IS element.

Step 3.

A BLASTN with ISFinder is applied on each IS sequence. If the e-value of the first hit is 0.0, then the ORF within this IS belongs to known (Real) IS already existing in the ISFinder database. Else, if the e-value is lower than $10^{-10}$, then we found a Partial IS.

Step 4.

Collect all Real IS from previous three methods (ALL_Tpase, Zigzag odd, and Zigzag even) and then remove overlaps among them. Finally, produce best Real IS with all information. Remark that the problem of finding consensus and overlaps can be treated as a lexical parsing problem.

4 The Proposed Pipeline

It is now possible to describe the proposed pipeline that can use the two modules detailed in the previous section. This pipeline, depicted in Figure 4, will increase the number of Real IS detected on the set of P.aeruginosa genomes under consideration (indeed, the detection is improved in all categories of insertion sequences, but we only focus on Real IS in the remainder of this article, for the sake of concision). Its steps are detailed in what follows.

Step 1: ORF identification.

Our pipeline is currently compatible with any type of annotation tools, having either functionality capability or not, but for comparison we only focus in this article on the following tools: BASys, Prokka, and Prodigal. BASys (Bacterial Annotation System) is a web server that performs automated, in-depth annotation of bacterial genomic (chromosomal and plasmid) sequences. It uses more than 30 programs to determine nearly 60 annotation subfields for each gene. Remark that genomes must be sent online manually, and that some curation stage may be required to remove some DNA ambiguity on returned genbank files.

Prokka (rapid prokaryotic genome annotation), for its part, is a classical command line software for fully annotating draft bacterial genomes, producing standards-compliant output files for further analysis [11]. Finally, Prodigal (Prokaryotic Dynamic Programming Genefinding Algorithm) is an accurate bacterial and archaeal genes finding software provided by the Oak Ridge National Laboratory [7].

Step 2: IS Prediction.

The second stage of the pipeline consists in using either NOASIS or DOASIS for predicting IS elements. Notice that NOASIS cannot be used with Prodigal, as this module requires information about gene functionality (both NOASIS and DOASIS can be use with Prokka and BASys annotations).

Step 3: IS Validation.

This step is realized by launching BLASTN on each predicted IS sequence with ISFinder. The e-value of the first hit is then checked: if it is 0.0, then the ORF within this sequence is a Real IS known by ISFinder. As described previously, it will be considered as Partial IS if its e-value is lower than $10^{-10}$. Both IS names of family and group are returned too.

5 Results and Discussion

We can firstly remark in Figure 5 that, using either Prokka or BASys for genes detection and functionality prediction is better than taking directly the annotated genomes from NCBI: a larger number of Real IS can be found. Additionally, this comparison shows that Prokka outperforms BASys in 3 families of ISs (namely: IS3, IS30, and ISNCY), while BASys seems better for detecting insertion sequences belonging in the IS5, IS1182, and TN3 families. This variability may be explained by the fact that functionality annotations of these tools depend probably on IS families that where known when these tools have been released.

The effects of DOASIS module compared to single OASIS on annotated NCBI genomes are depicted in Figure 6. The improvement in real IS discovery is obvious, illustrating the low quality and inadequacy of NCBI annotations for studying insertion sequences in bacterial genomes, and the improvements when using our pipeline. This chart shows too that a zigzag path in the annotation can oddly improve the detection of insertion sequences.

The prediction of real ISs is based on finding conserved regions (i.e., inverted repeats (IRs)) surrounded by transposase genes. Some ISs have been lost in All_Tpase, for the following reason: when we suggested that all genes are transposases, OASIS found predicted ISs that consist of large sets of transposases surrounded by IR in their left and right boundaries. But when these predicted ISs have been verified using ISFinder database, we did not find any good match. Contrarily, in Zigzag methods, good matches have been found (real ISs), because many of these elements consist of one or two transposase genes flanked by IRs. These results are listed with detail in Table 4 using BASys annotation tools.

We can thus wonder if the source of a wrong prediction of real IS is due to a wrong coding sequence prediction, or to functionality errors. Switching between NOASIS and DOASIS allows us to answer this question. We can conclude from Table 5 that (1) annotation errors are more frequent on NCBI, while Prokka annotates well the sequences related to ISs (see NOASIS columns), and that (2) both NCBI and Prokka have a better coding sequence prediction than BASys, at least when considering sequences involved in IS elements (see DOASIS columns and the correlation line). More precisely, the correlation is based on the number of predicted real IS elements between NOASIS and DOASIS.

Prodigal has been studied separately, as it does not provide genes functionality. The number of Real ISs per genome returned by our pipeline using prodigal is given in Figure 7. As shown in Table 6, the quality of coding sequences predicted with prodigal compared with other annotation tools allows us to discover the best number of real ISs. In particular, we have improved a lot of results produced by OASIS and ISFinder on NCBI annotations, which is usually used in the literature that focuses on bacterial insertion sequences. Furthermore, this table illustrates a certain sensitivity of coding sequence prediction tools with functionality annotation capabilities to detect ISs in some specific genomes like PA7. Indeed we discovered, during other studies we realized on this set of Pseudomonas strains, that PA7 has a lot of specific genes, that is, which are not in the core genome of all Pseudomonases, which may explain such a sensitivity.

6 Conclusion

Insertion sequences of bacterial genomes are usually studied using OASIS and ISFinder on NCBI annotations. We have shown in this article that a pipeline can be designed to improve the accuracy of IS detection and classification by improving the coding sequence prediction stage, and by considering a priori each sequence as a transposase. The source code for this pipeline can be download from the link 333http://members.femto-st.fr/christophe-guyeux/en/insertion-sequences. A comparison has been conducted on a set of Pseudomonas aeruginosa, showing an obvious improvement in the detection of insertion sequences for some particular configurations of our pipeline.

In future work, we intend to enlarge the number of coding sequence and functionality prediction tools and to merge all the Real IS results in order to improve again the accuracy of our pipeline. We will then focus on the impact of IS elements in P.aeruginosa evolution, comparing the phylogenetic tree of strains of this species with a phylogeny of their insertion sequences. Insertion events will then be investigated, and related to genomes rearrangements found in this collection of strains. We will finally enlarge our pipeline to eukariotic genomes and to other kind of transposable elements.