Ron Wehrens
Centro Ricerca e Innovazione, Fondazione Edmund Mach
Istituto Agrario di
San Michele all'Adige, Italy
Kurt
Varmuza
Ron Wehrens
Seminar lecture by Ron Wehrens*
at Institute of Statistics and Probability
Theory,
Vienna
University of Technology (invited by Peter Filzmoser)
17 November
2010
Biomarker identification in metabolomics
Abstract
Knowledge on which metabolites are
indicative of class differences is important for a better understanding of
underlying biological processes. To achieve this, several approaches can be
used. Among the most popular are the strategies based on PLSDA, either by
assessing the absolute size of the model coefficients or by looking at derived
statistics such as the variable-importance measure. These necessitate some form
of validation, in order to determine the optimal number of latent variables. Univariate statistics based on t-tests are potentially
simpler to apply but still need appropriate cut-off points, and cannot use
correlation information.
We discuss an alternative strategy for
biomarker discovery, based on the stability of the candidate biomarker set
under perturbance of the data. We evaluate the
performance of these using a real LCMS data set obtained by spiking apple
extracts with known metabolites at different concentrations, in order to mimic
real-life experimental conditions.
The scope of the data has been extended by
using these apple data for simulations of much larger data sets with similar
characteristics. In this way, we can assess the effect of the
variable-to-sample ratio on the biomarker identification, which in metabolomics usually is quite large.
This
is based on joint work with Pietro Franceschi, Urska Vrhovsek and Fulvio Mattivi.
New book
will be published soon <link>:
To Picture
Menu Last update: 2010-11-25 Copyright: K. Varmuza (Vienna, Austria) |