Marc Phander
A Neural Network Approach to Parameter Evaluation in Systems Biology
Bachelor Thesis in Theoretical Physics
Abstract:
This paper presents a set of tools and techniques originating
in the framework of artificial neural networks, that can be
used to investigate parameters for simulations of mathematical
models in systems biology. The central tool is ensemble of
multilayer perceptrons used as a classifier that generates a
prediction on the outcome of a simulation run with a given set
of parameters. A model for a gene regulatory network in the
shoot apical meristem of arabidopsis thaliana is used to
demonstrate possible difficulties for the training of the
classifier that can originate in the model itself as well as
in the algorithm used to search parameter space for good
configurations. A set of solutions for these difficulties is
presented and tested on data from the simulation, and it is
shown that these methods are sufficient to extract information
about the model and to give an accurate prediction if certain
conditions are fullfilled.
LU TP 08-16