Many top down approaches have been developed in recent years for network reconstruction from experimental data. Usually microarray data are used for this purpose, which allow for measuring a large number of gene product concentrations simultaneously. Hence, such datasets typically consist of several thousand gene products measured under a few conditions, leading to ill-posed optimization problems. The parameter space has to be restricted in a biologically meaningful way in order to obtain reliable results.
Accordingly, additional information about the system has to be included into the inference approach. Bayesian approaches provide a natural framework to combine different kinds of information about the system into one single framework. Here, experimental observations are considered as samples drawn from probability distributions, accounting for the stochastic nature of cell processes and noisy datasets. Prior knowledge is described by prior distributions over model parameters, and an investigation of the posterior distribution gives information about the posterior probability of model parameters after having seen the data.
Together with Dr. Lars Kaderali (Viroquant Research Group Modeling, University of Heidelberg), I have combined a Bayesian learning approach and a differential equation model to infer gene regulatory networks from gene expression time series data. In particular, a special prior distribution over interaction strengths favours sparse networks in which most of the strengths are small. We demonstrated that the method outperforms the classical maximum likelihood estimator in case of sparse and noisy datasets.
Radde N., Kaderali L., 2007. Bayesian inference of gene regulatory networks using gene expression time series data.
Lecture Notes in Bioinformatics (LNBI) 4414, Bird07, Springer Verlag, 1-15.
Kaderali L., Radde N., 2007. Inferring Gene Regulatory Networks from Gene Expression Data. In 'Computational Intelligence in Bioinformatics', Chapter 2, Studies in Computational Intelligence (SCI) series, Springer-Verlag, Berlin. [Abstract]
Radde, N., Kaderali L., 2007. A Bayes Regularized ODE Model for the Inference of Gene Regulatory Networks. submitted.