|
Abstract:
|
The past few years have seen a remarkable increase of interest in gene regulatory network research, by means of both biological and computational experiments. As for the latter, exploring the relationship between gene expression patterns and their genomic DNA sequences has special importance. We applied Bayesian logistic regression to model this relationship. A Metropolis-Hastings sampling scheme with variable selection is employed to obtain posterior samples of gene expression patterns given their sequence information, which is encoded according to predicted occurrences of over-representing motifs. This model not only provides comparable predict accuracy as other methods, but also suggests a way to improve motif discovery. Furthermore, the posterior samples can be used to evaluate measures of uncertainty of the predicted gene expression level under various conditions.
|
