Efficient implementations of Bayesian methods for cell signaling pathway model selection

Abstract

Cell signaling pathways are composed of chains of chemical reactions that propagate information within a cell. The mathematical modeling of those pathways is a relevant tool for testing their dynamics upon different initial conditions. In general, the modeling process starts with the definition of a signaling pathway cut for the mathematical model designing. In the sequence, an ordinary differential equation (ODE) system that describes the dynamics of each chemical species in the signaling pathway cut is calibrated with experimental observations of reactants and/or products from one or more reactions presented in such cut. One shortcoming in that approach is the huge space of alternatives to define a cut, rendering necessary the parameter estimation for a large number of models, a procedure very expensive, if not unfeasible, from the computational point of view. Therefore, we propose with this project the efficient implementation of parameter inference methods for cell signaling pathway cuts. To this end, we would focus on iterative Bayesian methods based on Kalman filters, for instance, the Extended Kalman Filter (EKF). We would implement those methods with C++ programming language and the OpenCL framework, possibly complemented with Julia programming language and CUDA computing architecture. The execution of our methods would take place in different platforms (e.g., in both CPU and GPU at once). We expect with this project to make a possible model selection of cell signaling cuts in many real-world problems, in particular in cancer biology studies. (AU)