Affiliations: Technische Universität Braunschweig, Institute of
Microbiology, Braunschweig, Germany | Massachusetts Institute of Technology, Department of
Chemical Engineering, Cambridge, MA 02140, USA | Technische Universität Braunschweig, Department of
Bioinformatics and Biochemistry, Braunschweig, Germany
Note: [] Corresponding author: Karsten Hiller, Massachusetts Institute of
Technology, Department of Chemical Engineering, 77 Massachusetts Ave., 56-439,
Cambridge, MA 02140, USA. Tel.: +1 617 2580 349; Fax: +1 617 258 687; E-mail:
[email protected]
Abstract: Modern high-throughput techniques allow for the identification and
quantification of hundreds of metabolites of a biological system which cover
central parts of the metabolome. Due to the amount and complexity of obtained
data there is an increasing need for the development of appropriate
computational interpretation methods. A novel data analysis pipeline designed
for high-throughput determined metabolomic data is presented. The combination
of principal component analysis (PCA) with emergent self-organizing maps (ESOM)
and hierarchical cluster analysis (HCA) algorithms is used to unravel the
structure underlying metabolomic data sets, including the detection of
outliers. Observed differences between various analyzed metabolomes are
automatically mapped and visualized using KEGG metabolic pathway maps. This way
typical metabolic biomarker for data sets from various analyzed growth
conditions and genetic backgrounds become visible. In order to validate the
described methods we analyzed time resolved metabolomic datasets obtained for
Corynebacterium glutamicum cells grown on various carbon sources
consisting of 126 different metabolic patterns. The analysis pipeline was
implemented in the user-friendly Java software eSOMet. The software was
successfully used for the clustering of the metabolome data mentioned above.
Metabolic biomarkers typical for the utilized carbon sources and analyzed
growth phases were identified.
Keywords: Emergent self-organizing maps, cluster analysis, comparative metabolome data analysis
