Modified variational method for genes regulatory network learning

We have revised the Markov lineal model used in the analysis of microarray time-series data. According to this model, the expression level of a given gene at any specific time is a linear combination of the measured expression levels of other genes at previous time instants, plus noise. The problem of uncovering such relationships can be solved using variational Bayesian methods. The linear model presented in the literature, however, establishes genetic relations between the data, which are assumed to have noise, whilst they should in fact be between the real expression levels. If this distinction is not taken into account, the noise is underestimated and the conclusions may not be valid. We have studied how the variational algorithm can be modified to solve this problem and propose a filteration to the linear model to include the real expression levels.