Bayesian hierarchical model for estimating gene association network from microarray data

Estimating gene association networks from gene microar- ray data is the key to decipher complicated Web of functional relationship between genes. However, the process remains to be challenging due to the relatively few independent samples and the large amount of correlation parameters. In a gene association network, vertices represent genes, and edges represent biological association between genes. The network edges are declared to be present if the corresponding correlation parameters are significantly different from a non-zero threshold. The approach has been very useful in inferring gene association networks, and facilitating network based discovery. However, as a Frequentist approach, it often suffers from the "overfitting" problem especially for analyzing small sample size data. Approaches that are able to globally estimate the correlation parameters with variance regularization followed by the seamless correlation thresholding are highly desirable.