Neural Network Models for Prediction of Steady-State and Dynamic Behavior of MAPK Cascade

The present work deals with the MAPK (mitogen-activated protein kinase), a three molecule module present in all eucaryotes, which has a wide range of functions in signal transduction, such as stress-response, cell-cycle-control, cell-wall-construction, osmosensing, growth and differentiation. This biological system is in fact an autonomous system and can be modeled by a set of ordinary differential equations. The aim is the construction of two computational models which predict the steady-state and dynamic behavior of proteins in the MAPK cascade. For the approximation of the steady-state stimulus/response behavior of proteins in the cascade a back-propagation neural network is used. The prediction of their dynamic behavior is a much more complicated and demanding task; 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 two neurons and it is trained by a dataset containing various initial conditions and the dynamical response of each protein. When the training process is complete, the appropriate weights are calculated and stored so as to produce a model which predicts the dynamic behavior of proteins in the cascade