Recurrent High Order Neural Networks for Identification of the EGFR Signaling Pathway

The present work deals with a specific signaling pathway called EGFR pathway (epidermal growth factor receptor) which is composed of twenty three proteins and their interactions. It is an essential part of the cell since it affects metabolism, growth and dimerization. The pathway can be modelled by an autonomous ODE. The aim is the construction of a computational model which predicts the dynamic behavior of each protein in the EGFR pathway. The mathematical tool used, is the so called recurrent high order neural network (RHONN). RHONN is a recurrent neural network with dynamical components distributed throughout its body in the form of dynamical neurons. It is applicable for the identification of dynamical systems. The RHONN model consists of twenty three neurons and it is trained by a set containing various initial conditions and the dynamical output of each protein. We use three different learning algorithms concluding to three different RHONN models. When the training stops the appropriate weights are calculated and frozen so as to produce reliable models to identify the EGFR pathway