Vol. 19 no.

17 2003, pages 2332–2333

BIOINFORMATICS APPLICATIONS NOTE DOI: 10.1093/bioinformatics/btg321

MetaGeneAlyse: analysis of integrated

transcriptional and metabolite data
Carsten O. Daub1, ∗ , Sebastian Kloska2 and Joachim Selbig1
1 Max
Planck Institute of Molecular Plant Physiology, Am Mühlenberg 1, 14476 Golm,
Germany and 2 Scienion AG, Volmerstraße 7a, 12489 Berlin, Germany

Received on March 26, 2003; revised on May 22, 2003; accepted on June 5, 2003

Downloaded from https://academic.oup.com/bioinformatics/article/19/17/2332/206374 by guest on 07 May 2021

ABSTRACT
Summary: New techniques in sample preparation allow high
throughput analysis of samples on the transcriptional as well
as on the metabolic level. We present a service accessible
via the web that allows the analysis of integrated data sets
that combine gene-expression data and metabolic data. After
uploading, data sets can be normalized, clustered by vari-
ous methods and results can be graphically visualized. All
calculations are carried out on a server, so even time- and
memory-consuming analyses can be done independently of
the performance of the client.
Availability: The service is accessible via web-interface at
http://metagenealyse.mpimp-golm.mpg.de/
Contact: daub@mpimp-golm.mpg.de
INTRODUCTION
Highly powerful experimental techniques allow large-scale
and parallel profiling of biological samples. Profiling on
the transcriptomic level is done using cDNA or oligonuc-
leotide hybridization arrays, which provide information on the
concentrations of mRNA (Baldwin et al., 1999). These micro-
array experiments are used to identify differentially expressed
genes, to make functional annotations of co-expressed genes
and to reconstruct regulatory networks. Recently, large-scale
analysis of metabolite levels became available which gives
additional insight into the internal state of the biochemical
system (Fiehn et al., 2000). Metabolic properties have been
used to distinguish different genotypes (Taylor et al., 2002).
Additionally, only integrated analysis of metabolites and tran-
scripts will reveal regulatory properties on a systems level
(Weckwerth and Fiehn, 2002; Weckwerth, 2003). Results
from both expression data and metabolite data can be used
to elucidate different levels of cellular regulation leading to
mechanistic insights into genetic and physiological control.
In an example, the identification of specific genes required
for the biosynthesis of a secondary metabolite that is used as
clinical agent showed the importance of the integrated analysis
(Askenazi et al., 2003).
∗ To whom correspondence should be addressed.
OVERVIEW
We have developed a web-based service which allows the ana-
lysis of integrated data sets containing gene-expression data
and metabolite data. At the web front-end, a tab-separated
data set can be uploaded. The file format is limited to an
ASCII text table containing genes and/or metabolites in rows
and experiments in columns. Data normalization is an essen-
tial preprocessing step in the comparison of transcriptional
and metabolic data, since both technologies provide different
measurement categories. We therefore implemented several
normalization algorithms like normalization to the maximum,
mean, median, variance, standard deviation, vector norm,
root mean square, z-score, and others. Analysing directly the
normalized data set can be done with k-means clustering and
principal component analysis. Different types of hierarchical
clustering, which are based on the prior calculation of a dist-
ance matrix, are also implemented. The underlying distance
matrix is then calculated as a first step using a distance measure
(e.g. Euclidean or Manhattan distance, Pearson’s correlation
coefficient or mutual information). The distance matrices can
also be downloaded to be post-processed by the user. Different
file formats for the download are available to import directly
the distance matrix into graph visualization tools. Analyses on
shuffled data sets are useful for the calculation of significance
levels, therefore our service allows the shuffling of uploaded
data sets. Documentation is available in PDF format and can
be downloaded from the web site.
IMPLEMENTATION
MetaGeneAlyse is a service that runs on a multiprocessor
Linux server under the web-server Apache. It consists of
a compilation of Perl scripts for the dynamic generation
of web pages and performance-optimized C++ programs
for time-consuming calculations. For the graphical visual-
ization of clustering results we use the statistics package R
(http://www.r-project.org/). The generated figures can be
displayed and downloaded in JPEG and PDF format. Time-
consuming calculations of distance matrices are handled by a
job queuing system, whereas smaller jobs are calculated on-
the-fly. This job management system allows the user to upload

2332 Bioinformatics 19(17) © Oxford University Press 2003; all rights reserved.

and analyse large data files containing up to 6000 genes and
metabolites.
Freely available tools for the exploration of microarray data
which run under Java on the user-side have been reported
(Dysvik and Jonassen, 2001; Sturn et al., 2002). An advantage
of our service is the calculation of large distance matrices
which gets computationally expensive and requires a large
working memory for larger data sets. Such data sets could not
be analysed with software running on client-side workstations.
The analysis of a data set consists of several steps from
normalization to the choice of a distance measure and a clust-
ering method. Each combination of these algorithms will tend
to emphasize different types of regulation inherent in the data.
The comparison of results obtained from different combina-
tions might give additional complementary information. For
instance, the results obtained by hierarchical clustering based
on different similarity measures like the Pearson correlation
coefficient and mutual information can be compared to detect
potential non-linear correlations in the data. Also, after dif-
ferent normalization steps, the same clustering algorithms can
be applied to analyse the impact of normalization to the data.
MetaGeneAlyse
REFERENCES
Askenazi,M., Driggers,E.M., Holtzmann,A.D., Norman,C.T.,
Iverson,S., Zimmer,P.D., Boers,M., Blomquist,R.P.,
Martinez,J.E., Monreal,W.A. et al., (2003) Integrated tran-
scriptional and metabolite profiles to direct the engineering of
lovastatin-producing fungal strains. Nat. Biotech., 21, 150–156.
Baldwin,D., Crane,V. and Rice,D. (1999) A comparison of gel-based,
nylon filter and microarray techniques to detect differential RNA
expression in plants. Curr. Opin. Plant Biol., 2, 96–103.
Dysvik,B. and Jonassen,I. (2001) J-express: exploring gene expres-
sion data using Java. Bioinformatics, 2, 96–103.
Downloaded from https://academic.oup.com/bioinformatics/article/19/17/2332/206374 by guest on 07 May 2021
Fiehn,O., Kopka,J., Dörmann,P., Altmann,T., Trethewey,N.R. and
Willmitzer,L. (2000) Metabolite profiling for plant functional
genomics. Nat. Biotech., 18, 1157–1161.
Sturn,A., Quackenbush,J. and Trajanoski,Z. (2002) Genesis: cluster
analysis of microarray data. Bioinformatics, 18, 207–208.
Taylor,J., King,R.D., Altmanm,T. and Fiehn,O. (2002) Application
of metabolomics to plant genotype discrimination using statistics
and machine learning. Bioinformatics, 18, 241S–248S.
Weckwerth,W. (2003) Metabolomics in systems biology. Annu. Rev.
Plant. Biol., 54, 669–689.
Weckwerth,W. and Fiehn,O. (2002) Patent PCT/EP03/00196.
2333