Estimation de paramètres et de variables cachées dans un modèle non linéaire à espace d'états pour l'inférence de réseaux biologiques

National audienceIdentifying biological networks requires to develop first, models able to capture the stochastic nature of biological processes as well as their dynamics and second, statistical learning methods to estimate their parameters from time-series measurements. Estimating quantitative nonlinear models of biological networks is still considered as a bottleneck for reverse engineering especially when the biological process is partially observed. We propose a general Bayesian framework based on Unscented Kalman Filtering that enables to estimate both parameters and hidden variables of nonlinear state-space model from data. In order to show the generality of our approach, we instantiate this framework on a transcriptional regulatory model based on Michaelis-Menten kinetics and a signaling pathway model based on mass action kinetics. Numerical results on parameter estimation from artificial data and real experimental data illustrate the approach. We showed that both parameters and hidden variables in biological networks models can be learned in the framework of nonlinear Kalman Filtering. Moreover a Bayesian approach allows directly to get uncertainty on parameters and hidden states. Eventually this method can be coupled with various structure inference methods used in Graphical Probabilistic Models