An optimized procedure greatly improves EST vector contamination removal
© Chen et al; licensee BioMed Central Ltd. 2007
Received: 24 January 2007
Accepted: 13 November 2007
Published: 13 November 2007
The enormous amount of sequence data available in the public domain database has been a gold mine for researchers exploring various themes in life sciences, and hence the quality of such data is of serious concern to researchers. Removal of vector contamination is one of the most significant operations to obtain accurate sequence data containing only a cDNA insert from the basecalls output by an automatic DNA sequencer. Popular bioinformatics programs to accomplish vector trimming include LUCY, cross_match and SeqClean.
In a recent study, where the program SeqClean was used to remove vector contamination from our test set of EST data compiled through various library construction systems, however, a significant number of errors remained after preliminary trimming. These errors were later almost completely corrected by simply using a re-linearized form of the cloning vector to compare against the target ESTs. The modified trimming procedure for SeqClean was also compared with the trimming efficiency of the other two popular programs, LUCY2, and cross_match. Using SeqClean with a re-linearized form of the cloning vector significantly surpassed the other two programs in all tested conditions, while the performance of the other two programs was not influenced by the modified procedure. Vector contamination in dbEST was also investigated in this study: 2203 out of the 48212 ESTs sampled from dbEST (2007-04-18 freeze) were found to match sequences in UNIVEC.
Vector contamination remains a serious concern to the data quality in the public sequence database nowadays. Based on the results presented here, we feel that our modified procedure with SeqClean should be recommended to all researchers for the task of vector removal from EST or genomic sequences.
The enormous amount of sequence data available in the public domain database has been a gold mine for researchers exploring various themes in life sciences. However, the quality of such data greatly affects subsequent analyses and hence is of serious concern to researchers. Data quality problems with the international sequence databanks have been reported previously [1, 2], for example, multiple representations of the same gene, vector contamination in a non-vector sequence [3, 4] and erroneous annotation of the exon/intron site in a gene .
Expressed Sequence Tags (ESTs)  are single reads of partially sequenced cDNA fragments. EST sequences have been applied to studies on gene prediction, alternative splicing, differential gene expression in specific tissues, etc.. More than half of the files in GenBank are EST data (based on release 154) , which were often produced in a batch sequencing project by automated processes. A computational processing procedure is therefore needed to obtain a reliable sequence which solely represents the molecular information of the cDNA. There is already a well-established procedure for EST sequence processing: trace data clean-up followed by removal of vector contamination. Various efficient bioinformatics tools such as phred [8, 9] have been made available on the internet for the purpose of trace data clean-up. Additionally, offline software like cross_match  by Phil Green, SeqClean  and LUCY  from TIGR , or the online program VecScreen  available at NCBI  are popular computer programs for vector contamination detection and/or removal.
Cross_match was implemented based on a restricted Smith-Waterman algorithm . The program executes a highly sensitive and efficient alignment to identify similarity between the two sequences under comparison, and it requires additional computer processing to remove the identified vector-like region from the target sequences. Cross_match has been incorporated into some EST processing packages such as PartiGene , a pipeline which integrates several bioinformatics programs to accomplish processes including sequence clean-up, EST clustering, assembly and functional annotation.
LUCY  searches for the greatest number of high quality subsequences in the query sequence according to the quality value of each base, which has been computed by a chromatogram base-calling program, such as phred [8, 9], on the trace data. LUCY then identifies the vector spliced or ligated to the input sequence based on a depth-first search algorithm, which gives a stringency of 95% minimal sequence similarity. The program finally removes the identified vector sequence from the input.
SeqClean  utilizes BLAST  to remove any sequence highly similar (minimum 94% identity by default) to a given list of vectors, adaptors, primers, or linker sequences that is located within 30% of total EST from the 3' or 5' end of the sequences to be trimmed. SeqClean also removes polyA repeats and applies low complexity filtering (in addition to performing sequence alignment) to identify similar vector segments in the target EST. Following the trimming process, the program "trashes" those sequences which are shorter than 100 nucleotides or contain more than 3% undetermined bases (N) in the resulting sequences.
Using BLAST, VecScreen can identify vector and primer contamination by comparing the input EST sequences at preset parameters against the Univec  database, which contains most of the commonly used commercial vectors and primers. However, several limitations are associated with this procedure. For example, a batch search function is not provided at the VecScreen website and furthermore, it does not remove those segments from the screened ESTs.
The SeqClean program was initially selected for use in our study. This program can be downloaded from the TIGR website for free, and it has been used in several EST projects [20–23] as well as a genome sequencing project . A large number of the EST sequences deposited in the public domain database may have been subjected to vector contamination removal by SeqClean. Here, we report an anomaly in the SeqClean trimming process and provide a refined procedure that greatly improves the trimming efficiency of SeqClean. Trimming efficiency was also compared to the other programs mentioned above. In addition, the error rate in the EST division of GenBank [3, 4] was evaluated.
Causes of errors in EST vector removal
Based on the observations above, a hypothesis was made that the trimming program neglected the circular form of the vector DNA which resulted in erroneous trimming of some of the EST sequences.
Most cloning vectors exist in a circular form in nature. However, to conveniently manage these molecules electronically, the sequences are usually displayed in linear form by opening the circular DNA at a specific position where the base is either numbered 1 (if located downstream) or the final base (if it is located upstream). Discontinuous numbering of the circular DNA sequence across the junction may lead to misinterpretation in bioinformatic analyses if the analysis program does not take this virtual transformation into consideration.
Marked improvement made to vector trimming with modified procedure
Effects of SeqClean on phred-treated EST against CVS or RVS form of the vector sequence
Number of ESTs to trim
Number of incompletely trimmed ESTs
% incomplete trimminga
Comparison of SeqClean, LUCY2 and Cross_match on CVS or RVS vectors
Cloning vectors, primers and adaptors used for ESTs in this study
Vector (cutting site)
Source of EST
primer1, primer2R, adapter1, adapter2R
primer1, primer2R, adapter1, adapter2R
Not I primer-adapter, Sal I adapter
Not I primer-adapter, Sal I adapter
trace archive at NCBI
Eco RI adaptors
trace archive at NCBI
Sfi-(dT)30; AAGCAGTGGTATCAACG CAGAGTGGCC
trace archive at NCBI
Effect of vector forms in different trimming programs
# resulting ESTsa
# removed ESTs
# incompletely trimmed ESTs
average length of trimmed ESTs
551.4 ± 163.4
560.7 ± 175.4
546.3 ± 176.7
539.7 ± 175.6
# resulting ESTsa
# removed ESTs
# incompletely trimmed ESTs
average length of trimmed ESTs
543.0 ± 172.1
563.5 ± 171.8
546.9 ± 176.7
540.1 ± 175.6
Among all the tested conditions (three programs and two vector forms), SeqClean trimming against RVS markedly outperformed all the other conditions, delivering the lowest error rate without removing any additional ESTs (Table 3 and Additional file 1 for details).
It appears that directly trimming the phred-cleaned sequences with LUCY2 gives a slightly better result than using quality files as input on all parameters tested (Table 3). Without modifying the vector form, LUCY2 in combination with phred cleaning appeared to surpass the other two programs, giving the lowest error rate without reducing the amount of useful information (Table 3). When trimming against RVSs, LUCY2 and cross_match did not seem to benefit from this modification as much as SeqClean did. It is thus proposed that the design flaw found with SeqClean very likely did not exist in the other two programs, hence rearrangement of the vector sequence may have little influence on these programs' performance.
Effect of vectors on vector trimming performance by three programs
The program leaves users to decide on which vector-free segments in the target sequence processed by cross-match should be output as the final product. For users who wish to conduct vector trimming with cross_match, software like PartiGene is recommended for it provides cleaned-up versions of the processed sequences. In this comparison of the trimming efficiency with the other two programs, we simply took the longest segment as the trimmed product.
Investigation of vector contamination rate of dbEST
As we observed different error rates caused by the vector trimming programs tested above, we were curious to see how much overall contamination remained in the dbEST . A total of 48212 entries was randomly sampled from the entire dbEST (2007-04-18 freeze) by retrieving every 600th EST entry from the database. These ESTs, were then subjected to BLAST analysis against UniVec database, which contains a redundancy-reduced set of representative sequences of commonly used cloning vectors. The same filtering criteria used above were applied to the BLAST results with E-value cutoff adjusted for the size of UniVec. 2203 (4.6%) of the sampled ESTs were observed to match some vector or primer/adaptor/linker sequence in Univec.
Vector contamination in ESTs sampled from dbEST
a# Contaminated ESTs
In summary, using vectors in CVS form, LUCY2 resulted in the lowest error rate among the three tested programs (Table 3). Though using the vectors in RVS form did not influence the trimming efficiency of LUCY2 or cross_match, RVS forms of the vectors allowed SeqClean to give the best trimming results of all tested programs (see Table 4). Thus we highly recommend our modified procedure to all researchers using SeqClean for the task of vector removal from EST or genomic sequences.
Questions over the quality of the sequence data deposited in the public domain database have caused great concern to life science researchers. Though dbEST provides VecScreen to help check the quality of submitted data, and the Gene indices at TIGR also filter EST data for further grouping, the quality of EST data is still not perfect, even in such a reputable data repository. Previous studies [3, 4] reported vector contamination rates of 0.28% to 0.36% for nucleotide sequences in GenBank. In this study, not only was significant vector contamination found in dbEST by analyzing randomly sampled EST data (see Results), but incomplete annotation of ESTs was also observed in both dbEST and Trace Archive. The former error, that is, presence of vector sequences in the ESTs, may cause assembly artifacts as well as errors in many other analyses depending on ESTs, while the latter error affects the amount of effective data for further mining.
This study re-examined the vector contamination rate in the data randomly sampled from dbEST. By looking at the 35363 ESTs (drawn at every 601st entry) which had been cloned into the 22 most prevalent vectors in the dbEST, 575 (1.63%) ESTs were still contaminated with vector sequences. One concern here was that the sampling procedure used here may result in an overrepresentation of the data submitted by very large sequencing projects. If data in such giant projects were incompletely trimmed, errors that existed in the data from other small projects may be overwhelmed and become invisible in the sampled data. In that case, an alternative sampling strategy should be considered. In our test dataset, the 35363 sampled ESTs used were submitted by more than 800 research groups, while the 575 contaminated ESTs were submitted by 100 of them. Furthermore, 174 (30%) of the 575 contaminated ESTs were "contributed" by one single group.
Our study here supports previous research  that contamination found in the GenBank data mainly existed in batches submitted by specific groups rather than being stochastically distributed (Additional file 3), suggesting the errors were derived from certain method(s). Yet, it can also be argued that the contamination may be more submitter-dependant than method-dependant. Poor data quality could be caused by various reasons including wet-laboratory practice and bioinformatics processing. However, no matter what the cause, any good vector trimming procedure ought to be able to remove as much contamination as possible.
An effective vector trimming procedure in addition to an efficient confirmation process are thus necessary prior to and during data submission to the public database, and are crucial factors in the reliability of further investigations. Though numerous well-known bioinformatics tools have been made freely available for specific data processing or analysis tasks, each has specific strengths and weaknesses.
An initial vector trimming task revealed a design flaw with the SeqClean program. The trimming errors with this program would occur if the size of the cDNA insert was so small that the start of the vector fell into the critical 30% range from either end of the EST sequences (Figure 1). Hence a vector like pTripleEx2, for example, whose MCS was located far from the start point (Table 2), would be unlikely to be affected by the design flaw of SeqClean. ESTs derived from a cDNA library of mostly long-insert clones may avoid encountering such errors altogether. This flaw was later almost completely overcome with the use of the modified vector form.
When errors are found with one program, an intuitively easy solution is to switch to a different program with similar features. However, there is always the possibility that any new program could give errors of a different type. The results presented in Tables 3 and 4 provide good examples.
The problem with SeqClean did not appear to exist in LUCY2 (Table 3), possibly due to the design of a user-provided splice file required by the program which specified the 100 to 150 bp vector sequences upstream or downstream of the ligation site. With the information in the splice file, the program can accurately anchor the vector sequence around the vector-insert border onto the ESTs in test. However, preparation of an accurate splice file is heavily reliant on users' thorough understanding of the cloning history and careful drafting of the border sequences.
Though cross_match gave a relatively higher error rate than the other two programs tested (Table 3), the length of the vector-matching segments remaining in the processed ESTs were all below 25 bases (Additional file 1). Furthermore, the trimming errors were distributed over various vectors. This implies that the incompletely trimmed sequences were more likely to result from the sensitivity of the algorithm than from an implementation error. Hence, it is probable that the trimming accuracy of cross_match could be improved by properly adjusting the parameter settings. As mentioned earlier, cross_match works in a relatively primitive manner despite its ease of use compared with the other two programs. cross_match identifies the vector-similar sequences in the EST without removing them, hence it requires further computer expertise to obtain a final result of trimmed sequences. Certain EST processing packages like PartiGene which incorporates cross_match for vector trimming in their multi-step pipeline would provide downstream processing to remove the unwanted vector-similar segments and to abandon the ESTs trimmed below a set length limit. Such a pipeline may be a good choice for users to accomplish multiple tasks of a specific purpose with one single procedure; however, it would be difficult to control the errors produced by individual programs along the pipeline. In a test with PartiGene using the same nine sets of EST data as the other programs, almost ten times as many ESTs (2014) were trimmed into <100 bases long segments (data not shown) and were finally removed, in comparison with the results shown in Table 3. Further investigation will be required to elucidate whether the atypical result generated by the program Partigene was due to an inherent design requirement of the processing pipeline, or whether it was caused by an implementation error in the post-processing scripts for cross_match alignments.
To correct an error found in a computer program, users with computer expertise may be able to modify the program code if the source code was made available. Nonetheless, fixing a computer program for re-implementation is often a tedious, uninteresting and error-prone job requiring extensive rewriting and debugging of someone else's code. Therefore we decided to take an alternative strategy by reforming in silico the linear form of the cloning vector with a simple word processing program like Notepad or Word. This allows the vector contamination to then be removed from the ESTs with SeqClean (against the reformed vector sequence). The resulting product is evaluated with BLAST.
Though only vector sequences are used for BLAST analysis, short appendix sequences like linkers, adaptors or primers should not affect the result of this confirmation process. As long as the boundaries of the vector sequence are correctly identified, the short sequences will be removed with the vector. On the other hand, if vector does get overlooked by the trimming program, the short sequences will remain untrimmed. This modified procedure for vector trimming clearly corrected almost all of the errors originally detected (Tables 1 and 3).
This procedure, which requires almost no program implementation, may be applicable to a lot of wet-lab based life-science research teams, who are interested in making use of the vast amount of EST data in the public domain but may lack extensive computer expertise. To simplify re-linearization of the cloning vector, a web-based DNA linearization program to pre-process the vector sequences was implemented, which is now available for free on the internet . BLAST evaluation of the performance of the vector trimming (Tables 2 and 3) confirmed that the vector re-linearization step turned SeqClean into an ideal tool for removal of vector contamination from automated sequencing data.
Therefore, we believe that the data quality in the public sequence databanks could be greatly improved if our modified procedure for vector removal was applied to data generated by the majority of high-throughput DNA sequencing projects prior to data submission.
Vector contamination remains a serious concern to the data quality in the public sequence database. This study identified that a modified procedure using RVS vector form could almost completely correct a design flaw found in SeqClean. The modified procedure, however, had little effect on the other two programs tested, LUCY2 and cross_match. When using vectors in CVS form, LUCY2 performed better than the other two programs under the same test conditions. SeqClean with RVS surpassed the other two trimming programs among all the tested conditions (CVS or RVS), and as it is so easy to use, SeqClean is highly recommended to all researchers for the task of vector removal from EST or genomic sequences.
EST sequences used in this work were acquired from three different resources. The EST sequences, with or without trace data available, used to study the effect of SeqClean in the initial and the final tests were derived from tomato cDNA clones prepared for research on the stress biology of plants by research laboratories at Agricultural Biotechnology Research Center, Academia Sinica (ABRC). Detailed molecular information about these sequences is stored at TSED (tomato stress EST database) . To compare the three computer programs for vector trimming, in addition to the ABRC ESTs, trace data in SCF format for ESTs were retrieved from Trace Archive at NCBI (June-12-2007 freeze) through the web interface. A total of 530,899 EST entries in Trace Archive  were found to contain all the required information for phred processing, including CHEMISTRY, CHEMISTRY_TYPE and CVECTOR. Finally, three cDNA libraries (ZM_BFC, GISZF001_ra and ISUM4) were selected for download. 3000 entries of the EST trace data were randomly drawn from each of the three cDNA libraries using a random number generator. To study the vector contamination rate with dbEST, the entire database of dbEST (2007/4/18 freeze), consisting of 42.3 million entries stored as 604 files, was firstly downloaded from the ftp server at NCBI and then parsed for the vector information recorded in each of the entries. ESTs were sampled at every 601st entry from the files of dbEST and a total of 48212 entries were collected. If an entry without vector information was encountered at sampling, it was removed immediately from the sampled collection (see below).
Around 14.4 million of the dbEST entries were found to contain no vector information and 118 of the 604 files were found to completely lack vector data. The vector information parsed out of dbEST was first grouped manually by synonyms. The number of ESTs under each vector was then counted. Around 74% of ESTs in dbEST were cloned into 22 vectors (Table 5). To investigate vector contamination by specific vectors, the EST entries with vector information among the 22 vectors were drawn from the 48212 sampled ESTs.
Vector trimming with SeqClean, LUCY2 or Cross_match and result evaluation
Those sequences whose trace data were available were firstly cleaned with the program phred prior to vector trimming by one of the three programs SeqClean, LUCY2 or cross_match. For phred cleaning, a -trim_alt option was included in the command line. Following vector trimming by the indicated bioinformatics program, the processed ESTs were subjected to BLAST analysis against the sequence of their cloning vector, as in Table 2. The highest scoring result for each BLAST analysis was parsed out for further filtering, with criteria based on the method described by Seluja and co-workers : (1) vector sequence occurs at either end of the EST, (2) the matched stretch is longer than 12 bases, and (3) shares cloning vector sequence identity of >95% for those EST pre-cleaned with Phred.
To examine the error rate in dbEST sequences from NCBI, the same criteria described above were applied to parse the BLAST output. Error rate of EST in this study indicates the ratio of ESTs which contained residual vector sequences (>12 base pairs) even after processing by a vector trimming program to the number of all the EST sequences subjected to the same trimming process.
Parameters were set at default values for SeqClean and LUCY2. For cross_match processing, a more sensitive setting (-minmatch 10 and -minscore 20) than default was applied, as recommended in the user manual.
For the LUCY2 trimming, in addition to files of EST or vector sequences, splice files were prepared as input data. A splice file contains two pairs of sequences: up to 150 base long sequences upstream or downstream of the cloning site on the vector in either orientation. Two splice files were prepared for each cloning vector, one for CVS and the other for RVS. The sequences in the splice files made from CVS or RVS of one cloning vector were basically identical except that the extended lengths derived from pDNR-LIB, pGEM-Teasy and pT7Blue were a little shorter for CVS than RVS since their splice sites fall within 100 bases from the start points of the vector sequences. The splice sequences for the three sets of ESTs derived from Trace Archive (i.e. those which were cloned into pCMV-SPORT6.1, pT7T3-PAC, and pDNR-LIB) were produced based on the library description documented in the dbEST. For the rest of the tested ESTs which were produced in our facility, the splice sequences were generated according to the commercial or the laboratory protocols for the cDNA library construction procedure and were later confirmed with the technician who performed the experiments. Prior to LUCY trimming, Bl2seq (a BLAST variant for pairwise sequence alignment) was used to double-check the generated splice sequences for the accuracy of the sequences and orientations.
The bioinformatics programs used in this study, including cross_match (version 0.990329), PartiGene (version 3.0.3), phred (version 0.020425.c), LUCY2 (version 2.19p-R07), SeqClean (downloaded on 07/13/2004), and BLASTN (version 2.2.15) were all obtained through their official websites. Large-scale data processing with phred, SeqClean, cross_match, PartiGene and BLAST programs was carried out at command line level on a Linux platform, while analysis with LUCY2 was performed on a Windows XP platform. Integration of the processed data and further analyses were performed with the programs Excel and Access on a Windows XP platform.
The web-based DNA linearization program was implemented in php language. Parsing of the trimming information was performed using a program written in perl. Batch BLAST was carried out with command lines on a Linux platform using the BLAST program downloaded from NCBI. The BLAST results were also parsed with an in-house perl program.
We want to thank Dr. Harry Wilson, Ms. M. Loney and Ms. Jou Lee for their careful editing of this manuscript. We are grateful to Mr. Dun-Hsiang Yang for invaluable discussion on statistical issues. We also want to thank Dr. Ning-Sun Yang and Dr. Na-Sheng Lin at the Agricultural Biotechnology Research Center, and Director Ker-Chao Li at the Institute of Statistical Sciences at Academia Sinica for supporting this study. Special thanks are given to the developers of SeqClean, phred, cross_match, LUCY2, PartiGene and BLAST programs.
- Bork P, Bairoch A: Go hunting in sequence databases but watch out for the traps. Trends Genet. 1996, 12: 425-427. 10.1016/0168-9525(96)60040-7.PubMedView ArticleGoogle Scholar
- Colleagues CTGoBMa: Quality control in databanks for molecular biology. Bioessays. 2000, 22 (11): 1024-1034. 10.1002/1521-1878(200011)22:11<1024::AID-BIES9>3.0.CO;2-W.View ArticleGoogle Scholar
- Seluja GA, Farmer A, McLeod M, Harger C, Schad PA: Establishing a method of vector contamination identification in database sequences. Bioinformatics. 1999, 15 (2): 106-110. 10.1093/bioinformatics/15.2.106.PubMedView ArticleGoogle Scholar
- Lamperti ED, Kittelberger JM, Smith TF, Villa-Komaroff L: Corruption of genomic databases with anomalous sequence. Nucleic Acids Res. 1992, 20 (11): 2741-2747. 10.1093/nar/20.11.2741.PubMed CentralPubMedView ArticleGoogle Scholar
- Korning PG, Hebsgaard SM, Rouze P, Brunak S: Cleaning theGenBank Arabidopsis thaliana data set. Nucleic Acids Res. 1996, 24 (2): 316-320. 10.1093/nar/24.2.316.PubMed CentralPubMedView ArticleGoogle Scholar
- Adams MD, Kelley JM, Gocayne JD, Dubnick M, Polymeropoulos MH, Xiao H, Merril CR, Wu A, Olde B, Moreno RF, Kerlavage AR, McCombie WR, Venter JC: Complementary DNA sequencing: expressed sequence tags and human genome project. Science. 1991, 252 (5013): 1651-1656. 10.1126/science.2047873.PubMedView ArticleGoogle Scholar
- GenBank release 154. [ftp://ftp.ncbi.nih.gov/genbank/release.notes/gb154.release.notes]
- Ewing B, Green P: Base-calling of automated sequencer traces using phred. II. Error probabilities. Genome Res. 1998, 8 (3): 186-194.PubMedView ArticleGoogle Scholar
- Ewing B, Hillier L, Wendl MC, Green P: Base-calling of automated sequencer traces using phred. I. Accuracy assessment. Genome Res. 1998, 8 (3): 175-185.PubMedView ArticleGoogle Scholar
- Cross_match. [http://www.phrap.org/]
- SeqClean. [http://www.tigr.org/tdb/tgi/software/]
- Chou HH, Holmes MH: DNA sequence quality trimming and vector removal. Bioinformatics. 2001, 17 (12): 1093-1104. 10.1093/bioinformatics/17.12.1093.PubMedView ArticleGoogle Scholar
- TIGR (The Institute for Genomic Research). [http://www.tigr.org/]
- VecScreen. [http://www.ncbi.nlm.nih.gov/VecScreen/VecScreen.html]
- NCBI (National Center for Biotechnology Information). [http://www.ncbi.nlm.nih.gov/]
- Smith TF, Waterman MS: Identification of common molecular subsequences. J Mol Biol. 1981, 147 (1): 195-197. 10.1016/0022-2836(81)90087-5.PubMedView ArticleGoogle Scholar
- Parkinson J, Anthony A, Wasmuth J, Schmid R, Hedley A, Blaxter M: PartiGene–constructing partial genomes. Bioinformatics. 2004, 20 (9): 1398-1404. 10.1093/bioinformatics/bth101.PubMedView ArticleGoogle Scholar
- Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ: Basic local alignment search tool. J Mol Biol. 1990, 215 (3): 403-410.PubMedView ArticleGoogle Scholar
- Univec. [http://www.ncbi.nlm.nih.gov/VecScreen/UniVec.html]
- Roeder V, Collen J, Rousvoal S, Corre E, Leblanc C, Boyen C: Identification of stress gene transcripts in Laminaria digitata (Phaeophyceae) protoplas cultures by expressed sequence tag analysis. Journal of Phycology. 2005, 41 (6): 1227-1235. 10.1111/j.1529-8817.2005.00150.x.View ArticleGoogle Scholar
- Brigido MM, Walter ME, Oliveira AG, Inoue MK, Anjos DS, Sandes EF, Gondim JJ, Carvalho MJ, Almeida NF, Felipe MS: Bioinformatics of the Paracoccidioides brasiliensis EST Project. Genet Mol Res. 2005, 4 (2): 203-215.PubMedGoogle Scholar
- Sterck L, Rombauts S, Jansson S, Sterky F, Rouze P, Van de Peer Y: EST data suggest that poplar is an ancient polyploid. New Phytol. 2005, 167 (1): 165-170. 10.1111/j.1469-8137.2005.01378.x.PubMedView ArticleGoogle Scholar
- Salmi ML, Bushart TJ, Stout SC, Roux SJ: Profile and analysis of gene expression changes during early development in germinating spores of Ceratopteris richardii. Plant Physiol. 2005, 138 (3): 1734-1745. 10.1104/pp.105.062851.PubMed CentralPubMedView ArticleGoogle Scholar
- Emrich SJ, Aluru S, Fu Y, Wen TJ, Narayanan M, Guo L, Ashlock DA, Schnable PS: A strategy for assembling the maize (Zea mays L.) genome. Bioinformatics. 2004, 20 (2): 140-147. 10.1093/bioinformatics/bth017.PubMedView ArticleGoogle Scholar
- Trace Archive. [http://www.ncbi.nlm.nih.gov/Traces/trace.cgi]
- TSED (tomato stress EST database). [http://abrc.sinica.edu.tw/tsed/]
- Boguski MS, Lowe TM, Tolstoshev CM: dbEST–database for "expressed sequence tags". Nat Genet. 1993, 4 (4): 332-333. 10.1038/ng0893-332.PubMedView ArticleGoogle Scholar
- Miller C, Gurd J, Brass A: A RAPID algorithm for sequence database comparisons: application to the identification of vector contamination in the EMBL databases. Bioinformatics. 1999, 15 (2): 111-121. 10.1093/bioinformatics/15.2.111.PubMedView ArticleGoogle Scholar
- DNA opener. [http://abrc.sinica.edu.tw/elcd/]
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.