Identifying latent dynamic components in biological systems

In systems biology, a general aim is to derive regulatory models from multivariate readouts, thereby generating predictions for novel experiments. However any model only approximates reality, leaving out details or regulations. These may be completely new entities such as microRNAs or metabolic fluxes which have a substantial contribution to the network structure and can be used to improve the model describing the regulatory system and thus produce meaningful results. In this poster, we consider the case where a given model fails to predict a set of observations with acceptable accuracy. In order to refine the model, we propose an algorithm for inferring additional upstream species that improve the prediction as well as the model fit and at the same time are subject to the model dynamics. In the studied context of ODE-based models, this means systematically extending the network by an additional latent dynamic variable. This variable is modeled by splines in order to easily access derivatives; the influence vector of the variable onto the species is then estimated from the data via model selection.