Mapping biological simulation results to phenotypes

OData support
Supervisor:
Dr. Antal Péter
Department of Measurement and Information Systems

In some biological models, like models of intracellular signaling networks, it is prob-

lematic to map the internal state of the model to phenotypes occurring in real experi-

ments. We have suspected that the signaling network model used by Turbine suers

from this problem. We have been using a simple manually built algorithm to extract

phenotypes from the model. I have tried to predict results of in vitro experiments from

the output of the underlying signaling network model with a recurrent neural

network. The neural network achieved mediocre results. The correlation between the

predicted and real experiment results is close to the correlation of results predicted

with the old algorithm, but it is lower. I have interpreted the weights learned by

the neural network with gene ontology analysis. The interpretation showed that the

neural network found some of the most important variables in the underlying model,

but it couldn't learn the temporal relationship between them. Nonetheless, the re-

sults are promising. With further development of the neural network I can probably

reach a better accuracy than the old algorithm.

Downloads

Please sign in to download the files of this thesis.