A Bayesian framework for knowledge driven regression model in micro-array data analysis

This paper addresses the sparse data problem in the linear regression model, namely the number of variables is significantly larger than the number of the data points for regression. We assume that in addition to the measured data points, the prior knowledge about the input variables may be provided in the form of pair wise similarity. We presented a full Bayesian framework to effectively exploit the similarity information of the input variables for linear regression. Empirical studies with gene expression data show that the regression errors can be reduced significantly by incorporating the similarity information derived from gene ontology.