W126–W131 Nucleic Acids Research, 2007, Vol. 35, Web Server issue
doi:10.1093/nar/gkm219
OPTIMIZER: a web server for optimizing the codon
usage of DNA sequences
Pere Puigbo` 1
, Eduard Guzma´ n1,2
, Antoni Romeu1
and Santiago Garcia-Vallve´ 1,
*
1
Evolutionary Genomics Group, Biochemistry and Biotechnology Department, Faculty of Chemistry, Rovira i Virgili
University (URV), c/MarcelÁli Domingo, s/n. Campus Sescelades, 43007 Tarragona, Spain and 2
Institut Catala` de la
Salut, A` rea Ba` sica de Salut, Tarragona 2, Spain
Received January 29, 2007; Revised March 22, 2007; Accepted March 28, 2007
ABSTRACT
OPTIMIZER is an on-line application that optimizes
the codon usage of a gene to increase its expression
level. Three methods of optimization are
available: the ‘one amino acid–one codon’ method,
a guided random method based on a Monte Carlo
algorithm, and a new method designed to maximize
the optimization with the fewest changes in the
query sequence. One of the main features of
OPTIMIZER is that it makes it possible to optimize
a DNA sequence using pre-computed codon
usage tables from a predicted group of highly
expressed genes from more than 150 prokaryotic
species under strong translational selection. These
groups of highly expressed genes have been
predicted using a new iterative algorithm. In addition,
users can use, as a reference set, a precomputed
table containing the mean codon usage
of ribosomal protein genes and, as a novelty, the
tRNA gene-copy numbers. OPTIMIZER is accessible
free of charge at http://genomes.urv.es/
OPTIMIZER.
INTRODUCTION
Gene expression levels depend on many factors, such as
promoter sequences and regulatory elements. One of the
most important factors is the adaptation of the codon
usage of the transcript gene to the typical codon usage
of the host (1). Therefore, highly expressed genes in
prokaryotic genomes under translational selection have a
pronounced codon usage bias. This is because they use a
small subset of codons that are recognized by the most
abundant tRNA species (2). The force that modulates this
codon adaptation is called translational selection and its
strength is important in fast-growing bacteria (3,4). If a
gene contains codons that are rarely used by the host,
its expression level will not be maximal. This may be one
of the limitations of heterologous protein expression (5)
and the development of DNA vaccines (6). A high number
of synthetic genes have been re-designed to increase their
expression level. The Synthetic Gene Database (SGDB)
(7) contains information from more than 200 published
experiments on synthetic genes. In the design process of a
nucleic acid sequence that will be inserted into a new host
to express a certain protein in large amounts, codon usage
optimization is usually one of the first steps (5). Codon
usage optimization basically involves altering the rare
codons in the target gene so that they more closely reflect
the codon usage of the host without modifying the amino
acid sequence of the encoded protein (5). The information
usually used for the optimization process is therefore the
DNA or protein sequence to be optimized and a codon
usage table (which we call the reference set) of the host.
Here we present a new web server, called OPTIMIZER,
for codon usage optimization focused on the heterologous,
or even homologous, gene expression in bacterial
hosts. OPTIMIZER allows three optimization methods
and uses several valuable, new reference sets.
OPTIMIZER can therefore be used to optimize the
expression level of a gene, to assess the adaptation of alien
genes inserted into a genome (8), or to design new genes
from protein sequences. The server is freely available at
http://genomes.urv.es/OPTIMIZER. It has been running
since July 2005 and it is updated twice a year with new
features and reference sets.
PROGRAM OVERVIEW
Implementation and input data
OPTIMIZER is an on-line application and its methods
are implemented in PHP (hypertext pre-processor)
programming language. The pre-calculated reference
tables are stored into a MySQL database. The data
input and the selection of the server options have been
organized in four steps. These steps are: (1) Input the
sequence to be optimized. DNA or protein sequences can
be used, although further steps are slightly different
depending on whether a DNA or protein sequence has
been input. (2) Input the reference set. Users can insert
*To whom correspondence should be addressed. Tel: þ34 977558778; Fax: þ34 977558232; Email: santi.garcia-vallve@urv.net
ß 2007 The Author(s)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/
by-nc/2.0/uk/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
a codon usage table in a variety of formats,
including tables from the Codon Usage Database (9),
or they can choose between 153 pre-computed codon
usage tables for ribosomal protein genes or a group of
highly expressed genes from prokaryotic genomes under
translational selection. Users can also choose a reference
set consisting of the tRNA gene-copy numbers.
(3) Choose the genetic code. (4) Choose the method
to be used in the optimization process. Depending on the
type of sequence introduced (DNA or protein) and
the reference set chosen, different optimization
methods are available (see below for a description of the
optimization methods).
Calculation of the reference sets
One of the main features of the OPTIMIZER server is
that it contains a series of pre-computed reference sets that
can be used in the optimization process. These reference
sets can be a table containing the codon usage of the host
(or the codon usage of a group of genes, such as the group
of highly expressed genes) or, as a novelty, the number of
tRNA gene copies predicted with the tRNA-scan software
(10). The pre-computed reference sets available in the
server are from more than 150 prokaryotic genomes that
are under a strong translational selection. The codon
usage reference tables available for these genomes
contain the mean codon usage of genes that encode
ribosomal proteins or a group of highly expressed genes.
Although the optimization process can be carried
out using the mean codon usage of the host organism as
a reference set, if the aim of the optimization process is
to increase the expression level of a gene, it is preferable
to use the codon usage of a group of highly
expressed genes. The mean codon usage of bacteria is
highly influenced by mutational bias (i.e. their G þ C
content). The optimal codons (those most frequently
used in highly expressed genes) are usually those that
agree with the mutational bias (i.e. G- or C-ending codons
for G þ C-rich organisms). However, the optimal codons
are not always in agreement with mutational bias.
For example, in the amino acids that are coded by
only two synonymous codons ending in C or T, the
C-ending codon is usually preferred, independently of
the mutational bias (3). Therefore, using the mean
codon usage of a genome may cause the wrong choice of
optimal codons.
A new feature of the OPTIMIZER server is that it can
use tRNA gene-copy numbers as a reference set for
the optimization process. If the codon usage bias of
highly expressed genes is caused by differences in tRNA
gene-copy numbers, why not use this information for the
optimization process? At present, information about
tRNA gene-copy numbers is used in the OPTIMIZER
server only with the ‘one amino acid–one codon’
optimization method (for a complete description of
the methods available, see the ‘Optimization methods’
section below).
Evaluation of which bacterial genomes are under
translational selection. Not all prokaryotic species are
under translational selection (4,11). It would be pointless
to optimize the codon usage of a gene in order to increase
its expression level in a species such as Helicobacter pylori,
which is not under translational selection (i.e. in which the
highly expressed genes do not have a different pattern of
codon usage from the other genes of their genome) (12).
Traditionally, correspondence analysis of the relative
synonymous codon usage of all genes from a genome
has been used to detect whether a genome is under
translational selection (13). In genomes under translational
selection, the ribosomal protein genes and other
highly expressed genes form a cluster in the correspondence
analysis plot, which confirms that highly expressed
genes have a different codon usage from the other genes
of a genome. This is the method we have used to detect
which bacterial species are under translational selection.
For each bacterial complete genome available, we made a
correspondence analysis using the Relative Synonymous
Codon Usage (RSCU) values of all the genes of a genome.
To automate the analysis of the correspondence plots, we
analyzed the position of the ribosomal protein genes
(expected to be highly expressed genes) along the four
principal axes obtained in the correspondence analysis.
If a genome is under translational selection, ribosomal
proteins and other highly expressed genes will show a
codon usage bias and they will form a cluster in the
correspondence plot. To make the prediction of translational
selection, we checked whether the mean position of
the ribosomal protein genes along any of the four
principal axes was significantly different (evaluated with
a t-test) from the mean position of the other genes of their
genome. To check our predictions, we also visually
inspected the correspondence plots (correspondence analysis
plots are available from the homepage of the server)
and analyzed the metabolic function of the predicted
highly expressed genes obtained. Analysis of 334 prokaryotic
genomes revealed that 153 genomes (the total
number of different species and genera was 108 and 63,
respectively) were under a strong translational selection.
These genomes were then used to calculate the precomputed
reference sets.
Prediction of highly expressed genes. The predicted highly
expressed genes were obtained using an iterative algorithm
that we have developed. This algorithm uses the group
of genes that encode ribosomal proteins as a seed and,
through a series of iterations, define a group of putative
highly expressed genes. This algorithm works as follows:
(i) Using the functional annotation, gene names or
COG families, genes that encode ribosomal proteins
are detected. Using the codon usage of these genes as
a reference set, the Codon Adaptation Index (CAI),
(14), at this stage namely CAIrp (15), is calculated
for each gene of a genome.
(ii) Using now the group of genes with the highest CAI
values as a reference set, the CAI for all genes is
recalculated. This process is repeated until a homogeneous
group is reached, i.e. when the group of
genes with the highest CAI values in one iteration is
the same as the group in the next iteration.
Nucleic Acids Research, 2007, Vol. 35, Web Server issue W127
To provide further support for our predictions, we
analyzed the metabolic functions of the putative highly
expressed genes. As expected, ribosomal proteins and
other expected highly expressed genes (16) were found
in the final group of predicted highly expressed genes.
To check our algorithm, we also analyzed species
not under translational selection. With these species,
either the algorithm never ended or the final group of
genes had a high codon usage bias but was not related
to their expression level. In this situation, neither
ribosomal protein genes nor genes expected to have a
high expression were included in the final group of
genes with a codon usage bias. Our method is similar to
the one developed by Carbone and co-workers (17).
However, these authors used all the genes of an organism
as the initial reference set, whereas we used ribosomal
protein genes.
Optimization methods
The OPTIMIZER server provides three methods for
optimizing the codon usage of the query sequence. In the
first method, the ‘one amino acid–one codon’ method, all
the codons that encode the same amino acid are
substituted by the most commonly used synonymous
codon in the reference set. However, this approach has
several drawbacks: for example, translational errors may
be made due to an imbalanced tRNA pool and it is
impossible to avoid repetitive elements or cleavage sites of
restriction enzymes (5,18). To overcome these drawbacks,
a second method, which we call the ‘guided random’
method, can be used. This method consists of a Monte
Carlo algorithm that selects codons at random based on
the frequencies of use of each codon in the reference set.
The third method, which we call the ‘customized one
amino acid–one codon’ method, is an intermediate
method in which users choose how many of the
59 codons (if the standard genetic code has been selected)
will be optimized with the ‘one amino acid–one codon’
approach. ‘Rare codons’ (i.e. the least used codons in the
reference set) are the first codons changed with this
approach. The aim of this third method is to maximize the
optimization by making the fewest changes in the query
sequence.
If the input sequence is a protein, it can be backtranslated
to DNA using the ‘one amino acid–one codon’
or the ‘guided random’ approach. If the ‘one amino
acid–one codon’ approach is selected, the protein
sequence can be back-translated to DNA using
codons with the highest G þ C or A þ T content,
or codons defined by Archetti (19) that minimize
mutation errors.
Outputs
Two indices, CAI and ENc (effective number of codons),
are used to measure the optimization process. CAI
measures the similarity between the codon usage of a
gene and the codon usage of a reference group of genes
(14). Its values range from 0 (when the codon usage of a
sequence and that of the reference set are very different) to
1 (when both codon usages are the same). This index is the
most effective of all codon bias measures for predicting
gene expression levels (12,20). The second index is ENc,
which is a measure of codon usage bias (21). Its values
range from 20 (if only one codon per amino acid is used)
to 61 (if all synonymous codons are used equally). Because
highly expressed genes usually use the minimal subset
of codons that are recognized by the most abundant
tRNA species, their ENc values are expected to be low.
Figure 1 shows some of the outputs provided by the
optimization of a DNA sequence: for example, the
query and optimized sequences and an alignment between
them, a chart of the relative frequencies of each codon of
the reference set and a codon usage table of the query
and optimized sequences. In addition, the OPTIMIZER
server has options for viewing or avoiding the cleavage
sites of the selected restriction enzymes (22) and
for splitting the optimized sequence into several
overlapping oligonucleotides for the construction of
a synthetic gene.
Comparison with other servers and programs
Table 1 shows a comparison of several public web servers
and stand-alone applications that allow some kind of
codon optimization. ‘GeneDesign’ (23), ‘Synthetic Gene
Designer’ (24) and ‘Gene Designer’ (18) are packages that
provide a platform for synthetic gene design, including a
codon optimization step. Other programs, such as
DNAWorks (25) and GeMS (26), focus more on the
process of oligonucleotide design for synthetic gene
construction. The stand-alone application INCA provides
an array of features, including now codon optimization,
which are useful for analyzing synonymous codon usage
in whole genomes (27). JCAT (28), ‘Codon optimizer’ (29),
UpGene (30) and the server presented here focus on the
codon optimization process. Although each server and
application has its own features, all of them have several
features in common. Most offer several options for the
input of the codon usage reference set. One of these
options is the possibility of using the tables from the
Codon Usage database (9). Usually, a limited number of
pre-computed tables of codon usage are available to be
used as a reference set in the optimization process.
In addition, not all of the available pre-computed
reference sets correspond to a group of highly expressed
genes (the proper reference set needed to optimize for
increasing gene expression level). Though most of the
programs and servers use a group of highly expressed
genes from E. coli as a pre-computed reference set, only
the ‘Synthetic Gene Designer’ and ‘GeneDesign’ servers
provide a pre-computed group of highly expressed genes
for 11 and 4 organisms, respectively. The exception is the
JCAT web server, which offers pre-computed tables
of predicted highly expressed genes from more than
200 bacterial species. However, this server uses the
method of Carbone et al. (17) to predict a group of
genes with a biased codon usage. These groups of genes
do not always correspond to a group of highly expressed
genes because not all bacterial species are under translational
selection (11,17). The high number of pre-computed
codon usage tables from bacteria and archaea that are
W128 Nucleic Acids Research, 2007, Vol. 35, Web Server issue
not under translational selection available in JCAT
therefore creates some confusion. The OPTIMIZER
server presented here provides the most pre-computed
codon usage tables for use as a reference set. The
OPTIMIZER server provides pre-computed tables for
more than 150 prokaryotic genomes that are under
strong translational selection. In addition, two groups of
genes are available in each reference set: a group of
highly expressed genes predicted using a new prediction
algorithm and the group of ribosomal protein genes.
OPTIMIZER is the only server or stand-alone application
that introduces a new kind of reference set such as
information about the number of copies of tRNA genes
for all the species included in the server. With regard to the
methods for codon usage optimization available in each
server or program, the first programs developed used only
the ‘one amino acid–one codon’ approach. More recent
programs and servers now include further methods to
create some codon usage variability. This variability
reflects the codon usage variability of natural highly
expressed genes and enables additional criteria to be
introduced (such as the avoidance of restriction sites)
in the optimization process. The OPTIMIZER server
presented here provides three methods of codon
optimization: a complete optimization of all codons,
an optimization based on the relative codon usage
frequencies of the reference set that uses a Monte Carlo
approach (similar to methods from other programs
and servers) and a novel approach designed to maximize
the optimization with the minimum changes between
the query and optimized sequences. Finally, note that
only the ‘Synthetic Gene Designer,’ INCA and
OPTIMIZER allow users to choose a non-standard
genetic code.
CONCLUSIONS
OPTIMIZER is a new codon optimization web server
focused on maximizing the gene expression level through
the optimization of codon usage. It has unique features,
such as a novel definition of a group of highly expressed
genes from more than 150 prokaryotic species under
translational selection, and the possibility of using
information on tRNA gene-copy numbers in the optimization
process. OPTIMIZER provides several precomputed
tables to specify a reference set and combines
Figure 1. Outputs provided from the optimization of a DNA sequence: (a) the optimized and query sequences and the indices (CAI, ENc
and%G þ C) for evaluating the optimization process, (b) codon usage tables of the query and optimized sequences, (c) query and optimized sequence
alignment to show changes in nucleotides (transitions or transversions) and (d) graphical view of the codon weight chart.
Nucleic Acids Research, 2007, Vol. 35, Web Server issue W129
three different methods of codon optimization.
The OPTIMIZER server can be used to optimize the
expression level of a gene in heterologous gene expression
or to design new genes that confer new metabolic
capabilities in a given species.
ACKNOWLEDGEMENTS
This work has been financed by project BIO2003-07672
of the Spanish Ministry of Science and Technology.
We thank Kevin Costello and John Bates of the Language
Service of the Rovira i Virgili University for their help in
writing the manuscript and two anonymous referees for
their helpful comments. Funding to pay the Open Access
publication charges for this article was provided by project
BIO2003-07672 of the Spanish Ministry of Science and
Technology.
Conflict of interest statement. None declared.
REFERENCES
1. Lithwick,G. and Margalit,H. (2003) Hierarchy of sequencedependent
features associated with prokaryotic translation.
Genome Res., 13, 2665–2673.
2. Ikemura,T. (1981) Correlation between the abundance of
Escherichia coli transfer RNAs and the occurrence of the respective
codons in its protein genes: a proposal for a synonymous codon
choice that is optimal for the E. coli translational system.
J. Mol. Biol., 151, 389–409.
3. Rocha,E.P. (2004) Codon usage bias from tRNA’s point of view:
redundancy, specialization, and efficient decoding for translation
optimization. Genome Res., 14, 2279–2286.
4. Sharp,P.M., Bailes,E., Grocock,R.J., Peden,J.F. and Sockett,R.E.
(2005) Variation in the strength of selected codon usage bias among
bacteria. Nucleic Acids Res., 33, 1141–1153.
5. Gustafsson,C., Govindarajan,S. and Minshull,J. (2004)
Codon bias and heterologous protein expression. Trends
Biotechnol., 22, 346–353.
6. Ivory,C. and Chadee,K. (2004) DNA vaccines: designing strategies
against parasitic infections. Genet. Vaccines Ther., 2, 17.
7. Wu,G., Zheng,Y., Qureshi,I., Zin,H.T., Beck,T., Bulka,B. and
Freeland,S.J. (2007) SGDB: a database of synthetic genes
Table 1. Comparison of OPTIMIZER with other similar freely available web servers and softwares
Name Methods Genetic code Reference set Reference
Web servers
OPTIMIZER – One amino acid–one codon Multiple – HEG from 4150 bacterial
genomes under TS
This article
– Guided Random (Monte Carlo
algorithm)b
– RPG
– Customized one amino
acid–one codon
– tGCN
– Codon usage database
– Defined by users
JCAT – One amino acid–one codon Standard – HEG from 4200 bacterial genomes 28
– Defined by users
Synthetic Gene
Designer (SGD)
– One amino acid–one codona
Multiple – HEG from six bacterial genomes 24
– Selective optimizationa
– Codon usage database
– Probabilistic optimizationa,b
– Defined by users
DNAWorks – Use of the two highest
frequency codons
Standard – HEG from E. coli 25
– Random – Codon usage tables for 10 species
– Codon usage database
– Defined by users
GeneDesign – One amino acid–one codon Standard – HEG from four species 23
– The next most optimal algorithm – Defined by users
– The most different algorithm
– Random
Stand-alone
applications
Gene Designer – One amino acid–one codon Standard – HEG from E. coli 18
– Monte Carlo algorithmb
– Codon usage tables for 25 species
– Codon usage database
– Defined by users
Codon optimizer – One amino acid–one codon Standard – HEG for several bacterial species 29
– Defined by users
INCA 2.1 – One amino acid–one codon Multiple – Mean codon usage of a whole genome or
selection of any group of genes
27
UPGene – One amino acid–one codon Standard – Eukaryotic, bacteria, yeast, plant and
worm predefined codon usage frequency tables
30
– Defined by users
GeMS – Monte Carlo algorithmb
Standard – Codon usage database 26
– Defined by users
Abbreviations used: HEG, codon usage of predicted highly expressed genes; RPG, codon usage of ribosomal protein genes; tGCN, tRNA gene-copy
number; TS, translational selection.
a
It uses an ‘optimality factor,’ defined as a scaling factor, to control the optimality of codon usage. Higher values of this factor mean low CAI values
and less optimized and more random codon usage.
b
These methods are essentially the same. They use the relative codon usage frequencies of the reference set as the relative probability that each codon
will be used in the optimization process.
W130 Nucleic Acids Research, 2007, Vol. 35, Web Server issue
re-designed for optimizing protein over-expression.
Nucleic Acids Res., 35, D76–D79.
8. Garcia-Vallve,S., Guzman,E., Montero,M.A. and Romeu,A. (2003)
HGT-DB: a database of putative horizontally transferred
genes in prokaryotic complete genomes. Nucleic Acids Res., 31,
187–189.
9. Nakamura,Y., Gojobori,T. and Ikemura,T. (2000) Codon usage
tabulated from international DNA sequence databases: status for
the year 2000. Nucleic Acids Res., 28, 292.
10. 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.
11. Willenbrock,H., Friis,C., Friis,A.S. and Ussery,D.W. (2006)
An environmental signature for 323 microbial genomes based on
codon adaptation indices. Genome Biol., 7, R114.
12. Henry,I. and Sharp,P.M. (2007) Predicting gene expression level
from codon usage bias. Mol. Biol. Evol., 24, 10–12.
13. Perriere,G. and Thioulouse,J. (2002) Use and misuse of
correspondence analysis in codon usage studies. Nucleic Acids Res.,
30, 4548–4555.
14. Sharp,P.M. and Li,W.H. (1987) The codon adaptation
index – a measure of directional synonymous codon usage
bias, and its potential applications. Nucleic Acids Res., 15,
1281–1295.
15. Nakamura,Y. and Tabata,S. (1997) Codon-anticodon assignment
and detection of codon usage trends in seven microbial genomes.
Microb. Comp. Genomics, 2, 299–312.
16. Karlin,S. and Mrazek,J. (2000) Predicted highly expressed genes
of diverse prokaryotic genomes. J. Bacteriol., 182, 5238–5250.
17. Carbone,A., Zinovyev,A. and Kepes,F. (2003) Codon adaptation
index as a measure of dominating codon bias. Bioinformatics, 19,
2005–2015.
18. Villalobos,A., Ness,J.E., Gustafsson,C., Minshull,J. and
Govindarajan,S. (2006) Gene designer: a synthetic biology tool for
constructing artificial DNA segments. BMC Bioinformatics, 7, 285.
19. Archetti,M. (2004) Selection on codon usage for error minimization
at the protein level. J. Mol. Evol., 59, 400–415.
20. Goetz,R.M. and Fuglsang,A. (2005) Correlation of codon bias
measures with mRNA levels: analysis of transcriptome
data from Escherichia coli. Biochem. Biophys. Res. Commun., 327,
4–7.
21. Wright,F. (1990) The ‘effective number of codons’ used in a gene.
Gene, 87, 23–29.
22. Roberts,R.J., Vincze,T., Posfai,J. and Macelis,D. (2005)
REBASE – restriction enzymes and DNA methyltransferases.
Nucleic Acids Res., 33, D230–D232.
23. Richardson,S.M., Wheelan,S.J., Yarrington,R.M. and Boeke,J.D.
(2006) GeneDesign: rapid, automated design of multikilobase
synthetic genes. Genome Res., 16, 550–556.
24. Wu,G., Bashir-Bello,N. and Freeland,S.J. (2006) The synthetic gene
designer: a flexible web platform to explore sequence manipulation
for heterologous expression. Protein Expr. Purif., 47, 441–445.
25. Hoover,D.M. and Lubkowski,J. (2002) DNAWorks: an
automated method for designing oligonucleotides for PCR-based
gene synthesis. Nucleic Acids Res., 30, e43.
26. Jayaraj,S., Reid,R. and Santi,D.V. (2005) GeMS: An advanced
software package for designing synthetic genes. Nucleic Acids Res.,
33, 3011–3016.
27. Supek,F. and Vlahovicek,K. (2004) INCA: Synonymous codon
usage analysis and clustering by means of self-organizing map.
Bioinformatics, 20, 2329–2330.
28. Grote,A., Hiller,K., Scheer,M., Munch,R., Nortemann,B.,
Hempel,D.C. and Jahn,D. (2005) JCat: a novel tool to adapt
codon usage of a target gene to its potential expression host.
Nucleic Acids Res., 33, W526–W531.
29. Fuglsang,A. (2003) Codon optimizer: a freeware tool for codon
optimization. Protein Expr. Purif., 31, 247–249.
30. Gao,W., Rzewski,A., Sun,H., Robbins,P.D. and Gambotto,A.
(2004) UpGene: Application of a web-based DNA codon
optimization algorithm. Biotechnol. Prog., 20, 443–448.
Nucleic Acids Research, 2007, Vol. 35, Web Server issue W131